A complete film.
A controlled process.

The guided creator path

Use this as the primary way through the studio. It is intentionally short. The workflow, handbook and checklists remain the detailed reference when a decision needs more depth.

The path at a glance

Before moving from idea to planning, apply the Shantanu story engine for character-driven narrative work: character, desire, obstacle, conflict and transformation. For informational, abstract or sensory work, define a deliberate progression of attention, information or feeling instead. For each key sequence, align camera, lighting, colour, music, pacing and intended emotional impact. Use the five-part director’s brief—role, objective, context, instructions and notes—for meaningful generation and handoff prompts.

The screen for every stage

The project workspace shows only one stage at a time. It asks for a short working note, applicable production units and a plain status. Expand the example or prompt only when useful. Use Mark this stage ready to record the creator's current decision. It does not bypass production checks or turn a draft into an approved film.

Every stage links to its detailed guide, the appropriate tool choices and its assigned checklist controls. A creator may loop between planning, making and review for every shot. They do not need to complete all shot planning before testing a high-risk shot; the next scene packet can incorporate the actual approved result.

A simple operating rule

Work from the smallest useful unit. Plan one scene, create one controlled shot, review its real result, then choose the next shot. When a result fails, identify the failure category and change the smallest relevant input. After two similar failures, change the method rather than repeating the same request.

Use the FABLE reference map when turning a plain-English idea into its first brief. It supplies the story card and feasibility questions; it does not prescribe a particular genre, tool or scene count.

Use the Shantanu masterclass reference to turn that brief into a dramatic engine, an emotional direction and a controlled prompt packet.

The workflow accommodates silent films, abstract imagery, single-shot work and deliberate ellipsis. Record why a conditional item does not apply. Never use an unchecked box, a successful job screen or an export notification as proof that the film is complete.

FABLE Bible: reusable story-to-production method

The supplied FABLE Competition Bible, Volume 2 is a creative reference collection containing twenty concept examples across grounded and anime styles. It is useful for how it turns a premise into an executable concept. It is not a source of verified platform capabilities, current tool pricing, eligibility rules or guaranteed outcomes.

What the Bible contributes

Add this to every new brief

Copy this lightweight card before developing a script:

FORMAT / AUDIENCE:
GENRE AND EMOTIONAL PROMISE:
LOGLINE:
TURN OR PAYOFF:
KEY VISUAL:
MOOD / CAMERA / SOUND RULES:
BEGINNING → CHANGE → CONSEQUENCE:
DELIVERY AND RIGHTS CONSTRAINTS:
RISKIEST SHOT OR EFFECT:
TEST AND RETRY BUDGET:

The card is a planning aid. A creator may decide that a film has no twist, that a key visual is deliberately absent, or that silence is part of the sound rule. Record that choice rather than forcing a familiar structure onto every genre.

Feasibility before generation

For each proposed beat, ask four questions before paying for a render:

1. What must the audience understand in this beat?
2. Which single visible action or change proves it?
3. What asset, motion, transformation, dialogue or sound dependency could fail?
4. What is the cheapest controlled test, and what will change if it fails twice?

If a concept depends on multiple changes, several points of view or a long chain of precise action, split it into shots and give each one a handoff. A polished mood description is not a substitute for physical action, continuity or edit coverage.

What is intentionally not adopted

The Bible’s named engines, directors, commercial categories, scene count, scene duration and statements such as “why it wins” are example-specific creative proposals. Their availability, pricing, limits, performance and eligibility must be checked at the time of use. The workflow never infers that a tool can complete a shot, that a concept will perform well, or that a competition accepts it.

Source boundary

This map is derived from the user-supplied PDF, FABLE Competition Bible — Volume 2. It summarizes a method and does not reproduce the collection’s full concepts, scripts or prose. Keep the original PDF with the project’s private source references.

The production workflow

Start with the guided creator path for a short, step-by-step route. Use this document as the decision and evidence reference behind each stage. Execution prompts: library · Higgs team case study: What we have done in the Higgs AI Film Festival · rationale: evidence.

Ownership and authority

Detailed role contracts, nineteen functions, scalable crew modes and the authoritative gate/RACI map are in Roles and handoffs. Agent dispatch and composition are defined in Multi-agent studio. Phase labels below describe contributors; the role map assigns the single accountable decision owner.

A solo creator may hold all roles, but must make each decision explicitly: producer owns scope/budget/delivery; director owns story and canon; generation operator owns composer accuracy and job records; continuity reviewer owns shot/state checks; editor owns joins and pacing; sound editor owns sync/mix; delivery reviewer owns the actual exported and uploaded artifact. Automation prepares evidence; it does not invent approval.

Keep a versioned brief, script, canon, asset registry, clip ledger, edit decision list and delivery ledger. Each status change records owner, timestamp, version and supporting evidence. User authorization persists within its stated model, cost and attempt limits; do not repeatedly ask for an already approved unchanged action.

References have scoped authority: the approved face controls identity, the location controls geography, and a selected approved boundary frame controls the incoming pose. Resolve conflicts before rendering. A negative sentence cannot reliably cancel a visible feature in a reference. An asset's existence in a project does not establish its attachment to a job.

The 31-area department playbook deepens each phase without expanding mandatory headcount. Compile selected department decisions into the timed scene packet. Use the filmmaker practice map for sourced inspiration and the specialist modules for concrete prompts and checks.

Adapt checks to the approved film

Use the production checklist guide and 97-check master alongside these phases. Repeat checks for every actual unit and version; reconcile inventory at each handoff. The HTML tracker supports the records but cannot discover missing media, verify evidence or confer approval.

Causality must be understandable according to the approved intent; it need not always be shown as literal on-screen mechanics. Deliberate offscreen action, ellipsis, animation and stylized physics are valid when planned and legible. Hold a clip for missing explicit contact/action only when that visibility is actually required by its brief. Character/prop/mechanism checks may be inapplicable to an empty environment, graphic or abstract shot, with a recorded reason. Select delivery mode at intake: platform publication, competition entry, or private/local handover. Upload and submission controls apply only to the relevant mode; actual final-file review and evidence-backed handover still apply.

Phases, gates and handoffs

State machines

Assets: exploratory → picked → approved → active → superseded/archived; rejected remains recoverable and cannot become continuity authority. Clips: planned → preflight-passed → submitted → processing → generated → reviewed → approved/repair/rejected → in-edit. Delivery: exported → file-verified → uploaded → platform-verified → published → submission-confirmed. Record these separately.

Timing and model decisions

Use short controlled units, commonly 3–5 seconds, where available and suitable. A project's approved longer takes override that default. Separate physical action time, requested render duration and retained edit duration. A normal-gravity short drop cannot be stretched to 15 seconds without changing its physics or using explicit slow motion; use a brief action and a hold, or trim. Do not copy historical 480p/1080p settings or credit prices as universal rules. Validate current controls before each spend.

Choose narration-first timing for presenter/explainer work; picture/performance-first for observational drama; score-led timing for intentional montage. Dialogue must be timed to an approved performance. Plan every destination's framing and caption safe areas; do not assume automatic cropping works. For a 2.39 crop inside 16:9, the retained vertical band is approximately 74.4%, centred unless editorially reframed.

Review and rating gate

The requested Divinium rubric is unresolved. The following is a proposed operational rubric, not a recovered or attributed rating system.

Hard gates: correct story outcome; approved identity/prop/geography; physically readable action; intended start/end state; usable picture/audio; required delivery compliance. Any failed hard gate blocks approval regardless of aesthetic score. Unknown is not pass.

Rate each dimension 0–4 with timestamp evidence: story clarity, continuity, physical performance, camera/editability, sound, technical quality. 0 unusable, 1 major repair, 2 material limitations, 3 usable as intended, 4 particularly strong. Proposed release threshold: every dimension ≥3 and all hard gates pass. “Not applicable” needs a reason; unheard audio stays unverified. Director may accept a minor deviation only with its visible effect and downstream implications recorded. Ratings apply to the retained interval and its joins, not merely a flattering still.

Review in passes: full-motion story; identity/geography; first/last frames and every cut; audio-only listening; uninterrupted cold viewing; technical decode/metadata. Sampling is useful triage but does not certify mechanics or listening. Finding one reference flash triggers an audit of all similar boundaries.

Completion and learning

Archive exact job metadata and approved boundary frames, source media, trims, edit project, stems, text/captions, master, alternate crops, provenance and submission receipt. Record defect → evidence → cause hypothesis → smallest change → tested result → scope where the lesson applies. Update the checklist at each real state change, never only at the end of a session.

Anti-drift change control and deadline triage

At intake configure a maximum attempt count and credit ceiling per shot, a protected edit/QA/upload time reserve, a picture-lock cutoff and a pivot checkpoint. Choose their values from the actual deadline and platform turnaround; never inherit them from a past film.

A change to identity, geography, a hero prop, causal action, dialogue or delivery format creates a new canon/card version and an impact list of dependent shots. Pause only those shots until the director resolves the change. A minor accepted visual deviation must be recorded with its allowed scope; it does not silently rewrite canon. Repeated failure after the authorized attempt limit stops retries, even when a new prompt sounds promising.

At each checkpoint estimate remaining critical-path generation, editing, review and upload time with uncertainty. If it exceeds time left after the protected delivery reserve, freeze optional polish and choose: simplify coverage, use a previously approved fallback, remove a nonessential beat with a coherent story revision, or pivot to a smaller complete story. A required causal beat cannot be removed merely to make the schedule appear green. If no compliant coherent cut can be delivered within the reserve, explicitly abandon/defer that entry rather than continue uncontrolled spending. Pivot requires a new brief and controlled assets, not reuse of incompatible canon.

Historical contrast only: the user reports abandoning The Apple at the Door after substantial drift shortly before deadline, then completing and submitting The Night Watch. Retrieved AatD defects support the drift diagnosis; Night Watch's later submission record supports completion. This does not prove a particular genre or clip length guarantees success.

Community working kit

The creator instruction manual

Start with Workflow; copy cards from Prompt library. Copy the blank production forms into your film project and follow the complete worked studio run. Historical examples and limits are in Evidence.

Prerequisites

Have a story outline, access to the intended generation account/project, an editor, a media player with frame stepping, storage for originals, and a defined budget. Optional ffprobe/ffmpeg can inspect media; a successful decode is not a substitute for viewing and listening. Record licensing/consent and destination requirements relevant to your material. Verify actual model controls; never assume controls exist because a template mentions them.

1. Set up the production

Start a fresh project using the checklist guide. Declare the actual asset/scene/shot/job/clip/join and delivery inventory, then use the complete checklist for each unit and revision. The companion HTML tracker records evidence, owners, drafts and stale checks; export its project JSON at each handoff. Counts do not replace gate decisions.

Before dispatch, select solo, small-crew or studio mode from Roles and handoffs. Record each function’s owner, backup, allowed actions and budget/publication limits in 00_admin/roles.md. Use Multi-agent studio when work is delegated. Roles do not create extra headcount or confer permission.

Create the following inside a new project, leaving original media unchanged:

00_admin/brief.md decisions.md budget.csv status.md
01_canon/canon_v001.md assets.csv continuity.csv
02_references/exploratory/ approved/ superseded/
03_story/script_v001.md shot_list.csv storyboards/
04_scenes/S001/packet.md prompts/ renders/ reviews/ boundaries/
05_audio/dialogue/ ambience/ foley/ score/ mix/
06_approved_clips/
07_edit/project/ proxies/ edl.csv
08_delivery/review/ masters/ platform_versions/ receipts/
09_archive/manifest.csv lessons.md

Name files PROJECT_S001_SH010_C01_v001_STATUS.ext. Never overwrite a source with a trim. Maintain one active-version pointer in status.md. An asset registry row contains ID, version, type, local path, platform project/ID, provenance, approved scope, excluded scope, status, approver and date. Clip rows add prompt version, references actually attached, model/mode, cost, job ID, source duration, usable in/out and review status.

2. Recover an existing project

List media, prompts, edit projects, review notes and submissions. Read dated task outcomes and job records. Distinguish local master, uploaded master and submitted master. Match by filename, duration, visual content and checksum where available. Reconcile stale notes with later evidence without deleting history. Record contradictory claims as unresolved until a stronger source resolves them. Do not start paid work merely to discover what already exists.

3. Write and lock the brief

Use P01. Write the audience promise, beginning/change/ending, runtime and delivery constraints. Include titles/credits/packshot time in the runtime budget. Separate assumptions from decisions. Use P02 for canon and P03 for the asset inventory. Pick the intended reference, generate on the required platform only if the project requires it, inspect the resulting image, then register its real ID. Local draft approval alone does not prove on-platform creation.

4. Prepare a shot

Use P04. For each scene, specify beginning state, emotional change and ending state. Divide into shots with framing (wide/medium/CU/ECU), camera side, primary action, sound and edit point. Design separate coverage for a setup, action and consequence if needed. A storyboard can guide planning without being attached; attach only frames that the chosen generation mode can use without producing a collage or reference flash.

Add the relevant department decisions

Use the department matrix and cards to assign specialist work within the existing nineteen functions. Copy the scene packet: its paired beat tables use the same timing to connect camera/focus/axis, light, action/contact, performance/gaze, dialogue, Foley/SFX, ambience, music/silence and edit boundaries. Choose only relevant M01–M31 modules, append them to P04/P05/P17/P19 as needed, and turn their checks into the task acceptance criteria. Resolve cross-department conflicts before preflight: for example, a camera move that hides the decisive hand contact must yield to the scene's causal purpose or receive separate coverage.

The Cameron/Spielberg practice map supplies source-backed inspiration and explicitly labelled AI adaptations. Apply playable emotional beats, legible geography, planned light/physical interaction and sound hierarchy to your genre. No director name belongs in the generation prompt as a shortcut. The worked run includes a filled timed example; Night Watch has its own restrained application.

5. Preflight the actual composer

Use P06 immediately before Generate:

• Confirm project/account, exact model and mode, duration, resolution, aspect ratio, audio, quality and batch.
• Compare the complete entered prompt, including its final line, with the approved text. Prefer paste over long simulated typing; re-read after entry.
• Open or inspect each selected reference thumbnail. Compare expected IDs/count against actual selected chips. A checkbox click is not proof of selection.
• Confirm that start-frame mode contains the correct start image, rather than an unrelated whole video or final frame.
• Record displayed cost and remaining authorized attempts. Stay within already approved limits.
• Submit once and capture job ID/status. If the UI fails, inspect existing jobs before trying again.

6. Review before continuing

Download the completed output and keep its job record. Play at normal speed with audio; compare critical mechanics frame by frame. Review both neighbours and first/last frames. Use P07 and the workflow rubric. Save the approved usable interval and a boundary frame at the actual chosen cut, not automatically the render's last frame. Record pose, gaze, held prop/hand, object count, camera side, screen direction, lighting and audio tail. A generated clip is not automatically approved.

7. Diagnose failures

8. Assemble and mix

Build a rough cut from approved intervals. Keep source-time and timeline-time in the EDL. Review causal story order; add editable titles/slates in the editor. Match exposure/colour across joins conservatively. Use continuous location ambience, synced Foley and a coherent separate score. Preserve the score's timing during picture-only repair unless the director approves changing it. Measure loudness/peaks against the destination's confirmed target and listen for clipping, sync, uneven levels and transitions. Do not interpret “contains an AAC stream” as “good sound.”

9. Deliver and close

Export a new version. Probe duration/streams; decode end to end; watch and listen through the actual file. Inspect all cards, crops, watermarks, packshots and cut boundaries, including replacements. Hash the approved master. For private/local handover, deliver that exact approved file to the named recipient or storage location and verify access and playback. For public release or a competition entry, upload that exact master and verify transcoded playback, visibility, title, thumbnail, captions and link; save the publication or submission confirmation separately. Archive originals, edit project, fonts/licensing notes, stems, EDL and receipts. Use P10 for lessons and update status immediately.

Worked generic example: The Last Parcel

Brief: a courier leaves a repaired toy at an empty workshop, revealing reconciliation without dialogue. Three-shot scene: wide arrival; CU parcel placement; reaction/empty doorway hold. Canon: one courier, blue coat, rectangular paper parcel with a single red cord, fixed left-hinged workshop door, overcast dawn. Sound: cloth, parcel contact, street air; score separate.

For the placement clip: approved hand/coat, parcel and bench references; incoming parcel held in the right hand; locked CU; one action places parcel on bench, fingers release, end on still parcel. Render duration is the supported setting chosen at preflight, with a brief hold after contact. Review cord count, parcel scale, contact/shadow, fingers, screen direction and clean head/tail. If the parcel duplicates, reject that interval and correct the reference/action block. If only the first few frames show a reference plate, trim them and recheck the arrival-to-placement join. Record the chosen boundary for the next shot. Never infer forgiveness through an unrequested smile or added dialogue.

Operating the approval and change log

For every gate G00–G19, record the accountable owner, candidate version, evidence, decision and affected next tasks. Use the gate register in ROLES_AND_HANDOFFS.md as the single ownership authority. On a material canon change, CC identifies affected descendants, AR retires old active pointers, DR approves the revised creative state, and PM reschedules within EP’s mandate. Pause only invalidated work. When a recipient cannot use an output, return a timecoded or field-specific defect and requested repair; do not restart an entire phase.

10. Use the instructor-derived sequence

For a new film, use P20 for capability/feasibility, P01 for the brief, P15/P02/P03 for asset options and canon, P16 for voice/performance, and P17 for timed boards. P04/P14 compile a packet and P05 supplies the media prompt. Only after P06's observed composer check and valid authority may a real production proceed. P07 reviews; P08 repairs; P19/P09 guide edit and delivery; P10 closes the session. The source map explains what was retained and corrected.

P18's reference experiments are optional research, not obligatory film production. Hold model, shot, action and settings fixed when comparing one attachment method; unlike scenes cannot isolate the cause of a better result. Explore options before identity lock; thereafter make views/state variants of that same identity. Do not regenerate every finalized asset merely because a historical lesson did: first verify whether the current project requires that provenance.

For Night Watch use the separate adaptation; its current submitted master remains preserved. For an unrelated film use the generic forms and variables, not Night Watch's cast, palette, score or runtime.

Using the production checklist

This checklist extends the Night Watch action plan and the existing AI Film Workflow checklist into the current studio workflow. It contains 97 checks covering all 31 specialist modules and all 20 decision gates. It is a reusable control set, not a promise that checking boxes produces a good film.

Start each film

1. Create a new project in the HTML tracker, with its own title and revision. The downloadable HTML contains no personal production state; the browser keeps entered state locally. Export the project JSON for durable handover and use Import to restore it on another machine. A browser's storage is not a backup or a shared team database.
2. Declare the actual inventory and boundaries. Add each asset, scene, shot, generation attempt, candidate clip, edit seam, change/repair, cut version, sound cue or scene mix, master, destination and archive package. No sampling: every unit needing a check gets its own record. At early stages some scopes are not yet known; keep inventory unconfirmed and record the missing set.
3. Assign named owners and pin current versions. The suggested role is responsibility guidance, not a person's identity. Use current approved canon and applicable destination rules; do not inherit Night Watch's deadline, runtime, team structure, safe band, score or model duration.
4. Work stage by stage using the workflow's G00-G19 gates. Each checklist row points to the relevant department modules and expected evidence. The gate register retains decision ownership; a tracker row does not authenticate authorization.
5. Update records when an actual state changes. Generated, reviewed, creatively accepted, exported, uploaded and submitted are distinct events. Record paths/IDs, timestamps or timecodes and review coverage; do not paste private account credentials.

What each state means

A checkbox is only a visual summary of a verified record. The HTML requires a named owner and evidence before verification; it requires an owner and reason for permitted exclusions. It records the current unit revision and date. These fields improve tracking but do not validate the truth of the evidence or replace independent review.

Repeat checks at the right granularity

Do not mark all scenes done because one passed. A rough-cut checklist cannot stand in for the exported master. Add replacement candidates as separate clip units. Update versions when relevant inputs change; the HTML marks previously verified checks stale for that unit. Cross-unit dependencies still require the producer's impact list and explicit invalidation of affected units; this tracker does not infer the dependency graph.

A check identity is its check ID plus unit ID and relevant revision. Record parent IDs and exact input/range/join context: scene to shot to attempt/output, and cut to ordered joins to master/destination. Update that context when selected in/out points, neighbours, mix or delivery requirements change, even if original bytes do not. The HTML invalidates verified rows for a changed unit; the producer must still update affected dependent units. Preserve failed/cancelled jobs and rejected unused clips with an explicit disposition and reason; they do not need to pass the current release's creative checks.

At every handoff, compare expected inventory with recorded units, then inspect failed/blocked/unverified/stale items and every required gate. The tracker reports counts, not an automatic release verdict. An inventory confirmation means the producer has reconciled the declared scope for the current review, not that a system independently discovered all media. A single pending required criterion keeps that criterion open regardless of overall percentage.

Every-session rhythm

Open: recover active versions; reconcile actual jobs/costs; add new units and resolve stale status. Before spend: finish that attempt's preflight and verify existing bounded authority. After a result: preserve source, review, record usable interval and actual cut boundary. Before continuing: check predecessor state and relevant departmental evidence. At edit/finish: sweep all joins and affected sound, then inspect the actual master. At close: export tracker JSON, save decisions and report completed portions and open gaps precisely.

Save, share and recover

Typing retains a draft; Save check applies the selected status after validation. Draft fields survive saving another row, changing filters and reloading when browser storage is available. Editing a verified or not-applicable record makes its previous verdict stale. Save changed unit context before verifying its checks. A saved draft is not an applied verification.

Keep one editable tab per browser. If another tab changes the saved projects, the older tab pauses saving to protect the newer data. Export its local changes before reloading; imports create separate project copies rather than silently merging decisions. Browsers without the safe storage-lock capability use export-only mode, and storage errors remain visible. Always export at handoff and before moving the HTML to another location; browser storage belongs to that browser and origin.

Share the blank HTML and PDF as reference material. Share a project JSON separately only when its owners and evidence are appropriate for the recipients. Browser printing excludes the interactive tracker, including hidden entries; the PDF contains only the reusable handbook and blank checklist. This review app has no cloud account, server database or live team synchronization.

What changed from Night Watch

The three preserved HTML checklists are historical snapshots. The later conversation and revised-master records show Night Watch was completed and submitted; pending scene boxes in older copies do not overturn that outcome. The enhanced checklist retains scene order, controlled references, signature shots, sound continuity, fallbacks and delivery checks. It adds evidence/version/owner tracking, repeated-unit inventory, full department coverage, bounded retries, actual-composer readback, boundary sweeps and distinct publication/submission/archive checks.

Independent review also clarified conditional checks for non-character/stylized shots, terminal failed jobs without output, separate source handles versus final runtime, repeated deadline checkpoints and local/private delivery without a festival submission.

The old fixed model durations, central safe band, continuous cello and competition runtime are project-specific. The old suggestion to stretch a falling key is replaced by physical timing and coherent coverage. A style-compliance tick is not evidence that the full prompt and reference chips reached the composer. Recovered later submission evidence is kept separate from present-day live platform state.

Review before packaging

This HTML/PDF edition is for the user's and AIC team's review. The existing local skill draft remains uninstalled. Skill finalization and the requested private GitHub project follow explicit user approval of this edition; public release would be a separate later decision. The exact Divinium rubric remains unidentified, so the working rubric is explicitly provisional.

The complete production checklist

Repeat checks for each actual unit. Capture check ID, unit ID/revision, status, named reviewer, evidence, date and next action. A verified tick is not authorization. Conditional exclusions need a reason.

Recovery and mandate

• C001 - Recover the latest project state (project; PM; G00; M17, M30). Inventory original media, approved versions, edit, uploads and receipts; resolve dated contradictions; do not import old ticks.
  Tools / Inspect and record: Files, backups and destination portals + Studio tracker and templates. Budget route: Files, backups and destination portals + LibreOffice. Use and trade-offs.
• C002 - Declare the complete affected set (project; AR; G00; M17). List all scenes, assets, shots, jobs, clips, joins, masters and destinations; record expected totals and known unknowns.
  Tools / Inspect and record: Files, backups and destination portals + Studio tracker and templates. Budget route: Files, backups and destination portals + LibreOffice. Use and trade-offs.
• C003 - Lock audience, premise and deliverable (project; EP; G01; M01, M02). Approved brief includes emotional promise, genre, ending, runtime including cards, technical destinations and exclusions.
  Tools / Record and decide: Studio tracker and templates + LibreOffice. Budget route: LibreOffice. Use and trade-offs.
• C004 - Assign owners and decision authority (project; PM; G01; M02). Named owners, backups, review independence and action/spend/publication limits; a role label is not authority.
  Tools / Record and decide: Studio tracker and templates + LibreOffice. Budget route: LibreOffice. Use and trade-offs.
• C005 - Set budget, retry scope and reserve (project; PM; G01; M02). Total and per-shot limits, cost units, cumulative prior attempts, queue uncertainty and protected edit/QA/upload reserve.
  Tools / Record and decide: Studio tracker and templates + LibreOffice. Budget route: LibreOffice. Use and trade-offs.
• C006 - Check current platform feasibility (project; ML; G01; M15). Dated actual controls, model/mode, duration options, reference/audio behaviour and costs; unsupported requests remain unknown.
  Tools / Compare and test: Higgsfield + Runway. Budget route: VicSee + ComfyUI. Use and trade-offs.
• C007 - Confirm destination and source-use requirements (project; LR; G01; M29). Current relevant rules, permissions, consent, credits and required disclosures; old contest conditions are not generic defaults.
  Tools / Human review: LibreOffice + Files, backups and destination portals. Budget route: LibreOffice. Use and trade-offs.
• C008 - Resolve physical-production applicability (project; PM; G01; M31). For hybrid filming, appropriate qualified human specialists and local process; mark purely synthetic work with a reason. Conditional; document applicability.
  Tools / Human expertise: LibreOffice + Blender. Budget route: LibreOffice. Use and trade-offs.

Story and creative direction

• C009 - Verify setup, action and consequence (project; WR; G02; M01). Beat map shows each required payoff and its visible or audible cause; ending matches audience promise.
  Tools / Write and review: LibreOffice + Studio tracker and templates. Budget route: LibreOffice. Use and trade-offs.
• C010 - Define emotional or experiential progression (project; DR; G02; M03). For narrative fiction: desire, obstacle, choice, subtext and observable performance change. For informational, abstract or sensory work: approved progression of attention, information or feeling, with the intended audience effect and review evidence. Conditional; document applicability.
  Tools / Write and review: LibreOffice + Studio tracker and templates. Budget route: LibreOffice. Use and trade-offs.
• C011 - Reconcile the runtime budget (project; ED; G02; M19). Final scene/shot durations plus retained cards/credits meet this brief. Source handles are budgeted separately and excluded from final runtime unless actually used.
  Tools / Write and review: LibreOffice + Studio tracker and templates. Budget route: LibreOffice. Use and trade-offs.
• C012 - Test the hardest dependency early (project; PM; G02; M02, M13). Identify high-risk identity/contact/physics/sync shot, smallest useful test and coherent fallback before dependent spend.
  Tools / Compare and test: Higgsfield + Runway. Budget route: VicSee + ComfyUI. Use and trade-offs.
• C013 - Approve a purposeful visual language (project; CV; G03; M04). Look bible records palette, contrast, texture, shape, recurring motifs and allowed variation without filmmaker-name shortcuts.
  Tools / Design and inspect: Higgsfield + Krita. Budget route: VicSee + Krita. Use and trade-offs.
• C014 - Approve the sound language (project; SD; G03; M20, M25). Voice, audible perspective, ambience, Foley/effect hierarchy, music policy and intentional silence serve this film.
  Tools / Spot and plan: DaVinci Resolve + Studio tracker and templates. Budget route: Audacity + LibreOffice. Use and trade-offs.

Canon and references

• C015 - Separate canon from allowed state changes (project; CC; G03; M14). Versioned identity, geography, props, appearance, lighting and sound; changes require scoped decisions.
  Tools / Record and decide: Studio tracker and templates + LibreOffice. Budget route: LibreOffice. Use and trade-offs.
• C016 - Select options before identity lock (project; DR; G03; M04, M07). Bounded exploration policy; after lock create views/states of the same identity rather than unrelated replacements.
  Tools / Design and inspect: Higgsfield + Krita. Budget route: VicSee + Krita. Use and trade-offs.
• C017 - Register exact source identity and provenance (asset; AR; G04; M17). Stable ID, revision/hash, local path, real platform project/ID if used, source/rights, lifecycle and approver.
  Tools / Inspect and record: Files, backups and destination portals + Studio tracker and templates. Budget route: Files, backups and destination portals + LibreOffice. Use and trade-offs.
• C018 - Check appearance and identity views (asset; CC; G04; M07). Useful face/body/detail views preserve garments, closures, hair, makeup and distinctive features under needed lighting. Conditional; document applicability.
  Tools / Design and inspect: Higgsfield + Krita. Budget route: VicSee + Krita. Use and trade-offs.
• C019 - Check environment geography (asset; PD; G04; M05). Entrances/exits, windows, landmarks, scale and camera zones agree across required location views. Conditional; document applicability.
  Tools / Plan and prove: Blender + Krita. Budget route: Krita. Use and trade-offs.
• C020 - Check prop design and mechanism (asset; PD; G04; M06). Exact count, shape, material, scale, handedness, moving parts and before/after states; reconcile conflicting image/text. Conditional; document applicability.
  Tools / Plan and prove: Blender + Krita. Budget route: Krita. Use and trade-offs.
• C021 - Declare reference influence and exclusions (asset; AR; G04; M15, M17). Each source controls specific identity/location/pose/style facts and must not control irrelevant state/background.
  Tools / Inspect and record: Files, backups and destination portals + Studio tracker and templates. Budget route: Files, backups and destination portals + LibreOffice. Use and trade-offs.
• C022 - Verify the real usable reference (asset; GO; G04; M16). Open actual file/thumbnail/handle in intended project; registration or click alone does not prove job attachment.
  Tools / Read back actual UI: Higgsfield + Files, backups and destination portals. Budget route: VicSee + Files, backups and destination portals. Use and trade-offs.
• C023 - Verify applicable permission and provenance (asset; LR; G04; M29). Evidence supports this asset's intended distribution and any identity/voice use; missing permissions stay unresolved.
  Tools / Human review: LibreOffice + Files, backups and destination portals. Budget route: LibreOffice. Use and trade-offs.

Scene planning

• C024 - State beginning, change and ending (scene; DR; G05; M01, M03). Scene intent and audience-information change are observable; required/excluded subjects and dialogue are explicit.
  Tools / Write and review: LibreOffice + Studio tracker and templates. Budget route: LibreOffice. Use and trade-offs.
• C025 - Pin incoming and outgoing states (scene; CC; G05; M14). Actual predecessor boundary where available; pose, gaze, hand, prop, source light and sound tails; no imagined approval.
  Tools / Watch and compare: CapCut + DaVinci Resolve. Budget route: Kdenlive. Use and trade-offs.
• C026 - Build coherent coverage and boards (scene; SB; G05; M13). Setup/action/consequence, exact panel count, source-clip splits, edit durations and handles; boards labelled planning-only or attachable.
  Tools / Plan and time: Blender + CapCut. Budget route: Krita + Kdenlive. Use and trade-offs.
• C027 - Plan framing and reveal order (scene; DP; G05; M08). Shot size, camera position/height, composition, visibility of critical action, crop safety and deliberate conceal/reveal.
  Tools / Plan and time: Blender + CapCut. Budget route: Krita + Kdenlive. Use and trade-offs.
• C028 - Plan field of view, perspective and focus (scene; DP; G05; M09). Viewpoint/near-far relationships, sharp region and focus target; numeric lens settings distinguished from prose intent.
  Tools / Plan and time: Blender + CapCut. Budget route: Krita + Kdenlive. Use and trade-offs.
• C029 - Plan blocking, axis and camera path (scene; DP; G05; M10). Actor path, eyeline targets, screen direction, movement start/end/pace; any axis crossing re-establishes geography.
  Tools / Plan and prove: Blender + Krita. Budget route: Krita. Use and trade-offs.
• C030 - Plan motivated light and state changes (scene; DP; G05; M11). Source direction/softness, fill, colour relationship, practical on/off, shadows and reflections carry across coverage.
  Tools / Plan and prove: Blender + Krita. Budget route: Krita. Use and trade-offs.
• C031 - Define exposure, motion and colour handling (scene; DP; G05; M12). Visible goals separate from measured fps/tags/transforms; no assumption that prompt words create log/HDR latitude.
  Tools / Inspect and measure: DaVinci Resolve + MediaInfo. Budget route: Kdenlive + MediaInfo. Use and trade-offs.
• C032 - Align set, props and appearance (scene; PD; G05; M05, M06, M07). Reference/version list matches script chronology, material states and physical contact needs; exclusions resolved.
  Tools / Plan and prove: Blender + Krita. Budget route: Krita. Use and trade-offs.
• C033 - Complete the shared timed sound plan (scene; SD; G05; M20). Same beat IDs/times as camera/light/action/performance/edit; dialogue, Foley, SFX, ambience, music and silence all considered.
  Tools / Spot and plan: DaVinci Resolve + Studio tracker and templates. Budget route: Audacity + LibreOffice. Use and trade-offs.
• C034 - Approve casting and voice performance (scene; DR; G05; M03, M21). Character/voice identity, exact words, pronunciation, accent, breath, pauses and delivery window; silence explicitly recorded. Conditional; document applicability.
  Tools / Record or generate voice: ElevenLabs · voice and SFX + Audacity. Budget route: Audacity. Use and trade-offs.
• C035 - Plan effects and repair dependencies (scene; FX; G05; M18). Required plates, masks, camera/light/contact/occlusion and protected regions; no nonexistent 3D/physics controls assumed. Conditional; document applicability.
  Tools / Plan effects dependencies: DaVinci Resolve + Blender. Budget route: Blender + Krita. Use and trade-offs.
• C036 - Set observable acceptance and review coverage (scene; QC; G05; M28). Critical tests, required viewing/listening/metadata, reviewer and next owner; no quality claim without appropriate evidence.
  Tools / Record and decide: Studio tracker and templates + LibreOffice. Budget route: LibreOffice. Use and trade-offs.

Shot and prompt readiness

• C037 - Compile a feasible shot prompt (shot; ML; G05; M15). Fixed canon and variable action/camera/light/audio separated; one compatible primary action; contradictions removed.
  Tools / Compare and test: Higgsfield + Runway. Budget route: VicSee + ComfyUI. Use and trade-offs.
• C038 - Prove physical timing and handoff feasibility (shot; CC; G05; M06, M14). Contact/support/release/consequence fit screen time; continuation starts at selected approved cut state.
  Tools / Plan and prove: Blender + Krita. Budget route: Krita. Use and trade-offs.
• C039 - Pin exact references and prompt versions (shot; AR; G05; M17). Expected IDs/files/hashes and attachment roles, start/end-image mode if supported, approved text including final line.
  Tools / Inspect and record: Files, backups and destination portals + Studio tracker and templates. Budget route: Files, backups and destination portals + LibreOffice. Use and trade-offs.
• C040 - Protect the needed edit interval (shot; ED; G05; M19). Planned usable range, handles, neighbour states and sound tail; causal action cannot be removed merely to fit.
  Tools / Watch and compare: CapCut + DaVinci Resolve. Budget route: Kdenlive. Use and trade-offs.

Every generation preflight

• C041 - Confirm account and destination project (attempt; GO; G06; M16). Observed intended identity/session/project and permitted action; do not silently switch accounts or browsers.
  Tools / Read back actual UI: Higgsfield + Files, backups and destination portals. Budget route: VicSee + Files, backups and destination portals. Use and trade-offs.
• C042 - Read back the whole entered prompt (attempt; GO; G06; M15, M16). Actual composer text matches approved version including final line; no truncated prefix accepted.
  Tools / Read back actual UI: Higgsfield + Files, backups and destination portals. Budget route: VicSee + Files, backups and destination portals. Use and trade-offs.
• C043 - Inspect actual selected attachments (attempt; GO; G06; M16). Expected count/IDs/thumbnails/mode match actual chips; no unrelated video, wrong frame or missing reference.
  Tools / Read back actual UI: Higgsfield + Files, backups and destination portals. Budget route: VicSee + Files, backups and destination portals. Use and trade-offs.
• C044 - Read back model and settings (attempt; GO; G06; M12, M16). Actual mode, model, duration, ratio, resolution, batch, audio and quality; only supported current values.
  Tools / Read back actual UI: Higgsfield + Files, backups and destination portals. Budget route: VicSee + Files, backups and destination portals. Use and trade-offs.
• C045 - Check valid authority and remaining limits (attempt; PM; G06; M02, M16). Displayed cost and reservation fit existing approval, cumulative shot attempts, cost units and delivery reserve.
  Tools / Read back actual UI: Higgsfield + Files, backups and destination portals. Budget route: VicSee + Files, backups and destination portals. Use and trade-offs.
• C046 - Submit once and reconcile the job (attempt; GO; G07; M16). Unique operation/job mapping, observed status and actual cost; uncertain response checked before retry.
  Tools / Read back actual UI: Higgsfield + Files, backups and destination portals. Budget route: VicSee + Files, backups and destination portals. Use and trade-offs.
• C047 - Preserve the actual result or terminal job record (attempt; AR; G07; M17). For a completed job preserve the readable original and exact metadata. For a failed/cancelled job retain its terminal evidence, actual charge/refund and no-output status. Uncertain jobs remain unresolved; none is creatively approved by this check.
  Tools / Inspect and record: Files, backups and destination portals + Studio tracker and templates. Budget route: Files, backups and destination portals + LibreOffice. Use and trade-offs.

Every clip review

• C048 - Review the actual full required modalities (clip; QC; G08; M28). Normal-speed motion, critical frame steps and listening as required; sampled images or a stream flag are not full review.
  Tools / Watch and listen: DaVinci Resolve + CapCut. Budget route: Kdenlive. Use and trade-offs.
• C049 - Check story and performance (clip; DR; G08; M03). Visible beat and emotion read; gaze/gesture/subtext and exact spoken meaning agree with approved intent.
  Tools / Watch and compare: CapCut + DaVinci Resolve. Budget route: Kdenlive. Use and trade-offs.
• C050 - Check identity and appearance continuity (clip; CC; G08; M07, M14). Face, body, wardrobe, hair, accessories and permitted state changes match pinned references throughout. Conditional; document applicability.
  Tools / Watch and compare: CapCut + DaVinci Resolve. Budget route: Kdenlive. Use and trade-offs.
• C051 - Check props, scale and geography (clip; CC; G08; M05, M06). Count, shape, relative size, material, doors/windows/landmarks, held hand and locations remain consistent. Conditional; document applicability.
  Tools / Watch and compare: CapCut + DaVinci Resolve. Budget route: Kdenlive. Use and trade-offs.
• C052 - Check contact, occlusion and physical causality (clip; FX; G08; M06, M18). Where required by the approved intent, check contact/support/release/occlusion and mechanism consequence. Approved stylization, offscreen action or ellipsis must preserve intended audience causality; do not impose realism or explicit mechanics universally. Conditional; document applicability.
  Tools / Watch and compare: CapCut + DaVinci Resolve. Budget route: Kdenlive. Use and trade-offs.
• C053 - Check camera, focus and spatial continuity (clip; DP; G08; M08, M09, M10). Critical detail readable, focus change not morph, stable landmarks and motivated motion/axis/eyelines.
  Tools / Watch and compare: CapCut + DaVinci Resolve. Budget route: Kdenlive. Use and trade-offs.
• C054 - Check light, exposure and motion quality (clip; DP; G08; M11, M12). Source states, shadows/reflections, colour/contrast, cadence and blur consistent; important detail preserved.
  Tools / Inspect and measure: DaVinci Resolve + MediaInfo. Budget route: Kdenlive + MediaInfo. Use and trade-offs.
• C055 - Check all audible content (clip; SD; G08; M20, M21, M22, M23, M24). Listen for correct words/voice, contact/sync, ambience, unwanted speech/music, clipping and usable tails; silence verified when intended.
  Tools / Listen and compare: DaVinci Resolve + REAPER. Budget route: Audacity + DaVinci Resolve. Use and trade-offs.
• C056 - Inspect the complete head, tail and edit range (clip; ED; G08; M19). No reference flash/collage/text/reset; selected interval and actual boundary usable with neighbours.
  Tools / Watch and compare: CapCut + DaVinci Resolve. Budget route: Kdenlive. Use and trade-offs.
• C057 - Record evidence and creative acceptance separately (clip; AR; G08; M17). Timecoded findings, coverage/unknowns, exact source/usable in-out, decision owner and accepted boundary; generation is not approval.
  Tools / Record and decide: Studio tracker and templates + LibreOffice. Budget route: LibreOffice. Use and trade-offs.

Every join

• C058 - Check both sides of the actual cut (seam; CC; G12; M14). Pose/gaze/hand/prop/source light and screen direction agree or have an approved motivated transition.
  Tools / Watch and compare: CapCut + DaVinci Resolve. Budget route: Kdenlive. Use and trade-offs.
• C059 - Check full motion and causal rhythm at the join (seam; ED; G12; M19). No duplicated action, reset, missing beat, frozen padding, flash, black gap or accidental frame.
  Tools / Watch and compare: CapCut + DaVinci Resolve. Budget route: Kdenlive. Use and trade-offs.
• C060 - Check acoustic continuity across the join (seam; SD; G13; M24, M26). Dialogue/contact sync, ambience perspective, score phrase, tails and silence fit without clicks or level jumps.
  Tools / Listen and compare: DaVinci Resolve + REAPER. Budget route: Audacity + DaVinci Resolve. Use and trade-offs.
• C061 - Sweep similar joins after a shared defect (seam; QC; G12; M28). Check all affected boundaries of the same failure class; one repaired sample does not close the set. Conditional; document applicability.
  Tools / Watch and compare: CapCut + DaVinci Resolve. Budget route: Kdenlive. Use and trade-offs.

Repair and change control

• C062 - Diagnose and propose the smallest repair (change; ML; G09; M15, M18). Exact time/defect, evidence, causal hypothesis and isolated prompt/reference/edit change; preserve unaffected approved work.
  Tools / Diagnose and propose: DaVinci Resolve + Blender. Budget route: Blender + Krita. Use and trade-offs.
• C063 - Retain cumulative retries and spend (change; PM; G09; M02). Same shot scope includes prior jobs and reservations; renamed tasks or model changes do not reset limits.
  Tools / Record and decide: Studio tracker and templates + LibreOffice. Budget route: LibreOffice. Use and trade-offs.
• C064 - Resolve canon change and dependent work (change; DR; G10; M14). Recorded decision, new revision, affected assets/shots/edits/sound/reviews; invalidate dependent acceptance only. Conditional; document applicability.
  Tools / Record and decide: Studio tracker and templates + LibreOffice. Budget route: LibreOffice. Use and trade-offs.
• C065 - Rerun affected checks on the replacement (change; QC; G10; M28). Review actual replacement and neighbours. Changes to source in/out, timeline order, join neighbours, timing, mix or delivery requirements invalidate relevant evidence even when source-file bytes are unchanged.
  Tools / Watch and listen: DaVinci Resolve + CapCut. Budget route: Kdenlive. Use and trade-offs.
• C066 - Re-evaluate continue, fallback, pivot or defer (checkpoint; EP; G11; M02). At each scheduled checkpoint or material time/cost/dependency change, record current critical path plus reserve, remaining limits and decision. Preserve prior checkpoint records. Fallback retains approved story causality, including deliberate ellipsis where intended.
  Tools / Record and decide: Studio tracker and templates + LibreOffice. Budget route: LibreOffice. Use and trade-offs.

Editorial and picture lock

• C067 - Assemble only accepted source intervals (cut; ED; G12; M19). Timeline/source-time mapping, versions, relinkable originals, trims and intentional transitions; missing coverage explicit.
  Tools / Watch and compare: CapCut + DaVinci Resolve. Budget route: Kdenlive. Use and trade-offs.
• C068 - Review the whole story and rhythm (cut; DR; G12; M01, M03, M19). Setup/reveal/payoff, performance pauses, scene order and ending work in the full cut, not only isolated clips.
  Tools / Watch and compare: CapCut + DaVinci Resolve. Budget route: Kdenlive. Use and trade-offs.
• C069 - Verify final VFX and protected regions (cut; FX; G12; M18). Full affected motion, scale/contact/occlusion/light, matte edges and neighbours; no unintended identity/background change. Conditional; document applicability.
  Tools / Inspect the final VFX: DaVinci Resolve + Blender. Budget route: Blender + Krita. Use and trade-offs.
• C070 - Control titles, graphics and captions (cut; ED; G12; M19). Editable accurate typography, duration, placement and safe areas; generated text artifacts removed or approved explicitly.
  Tools / Edit and human review: CapCut + DaVinci Resolve. Budget route: Kdenlive + Shutter Encoder. Use and trade-offs.
• C071 - Approve a named picture-lock version (cut; DR; G12; M19). Exact timeline/version, remaining permitted finish changes and accountable human decision; timing changes reopen dependent sound/captions.
  Tools / Record and decide: Studio tracker and templates + LibreOffice. Budget route: LibreOffice. Use and trade-offs.

Sound and music

• C072 - Spot the audible story priority (sound; SD; G13; M20). Cue/time/owner, listener perspective, dominant element and deliberate silence; all picture changes reconciled.
  Tools / Spot and plan: DaVinci Resolve + Studio tracker and templates. Budget route: Audacity + LibreOffice. Use and trade-offs.
• C073 - Check dialogue, voice and ADR (sound; SD; G13; M21). Exact script, pronunciation/voice/consent, breaths, no clipped words, natural timing and lip/action sync. Conditional; document applicability.
  Tools / Record or generate voice: ElevenLabs · voice and SFX + Audacity. Budget route: Audacity + Kdenlive. Use and trade-offs.
• C074 - Check Foley against visible contact (sound; SD; G13; M22). Right surface/material/weight/distance, contact/release frames, no extra steps or repeated impacts. Conditional; document applicability.
  Tools / Perform and synchronize: Audacity + REAPER. Budget route: Audacity + Kdenlive. Use and trade-offs.
• C075 - Check designed effects and mechanisms (sound; SD; G13; M23). Cause, attack/body/tail, recurring identity, timing and perspective; no music/speech contamination or masked critical cue. Conditional; document applicability.
  Tools / Design and synchronize: ElevenLabs · voice and SFX + REAPER. Budget route: Freesound + Audacity + Kdenlive. Use and trade-offs.
• C076 - Check ambience and room tone (sound; SD; G13; M24). Continuous beds/patches and location perspective, clean loops/tails, no unwanted voices or impossible environmental sound. Conditional; document applicability.
  Tools / Build and listen: REAPER + DaVinci Resolve. Budget route: Audacity + Freesound + Kdenlive. Use and trade-offs.
• C077 - Check music purpose, timing and final status (sound; SD; G13; M25). Intentional motif/entry/release/silence; approved score timing protected; temp replaced/cleared and rights documented. Conditional; document applicability.
  Tools / Compose, clear and edit: Suno + REAPER. Budget route: Audacity + Freesound + Kdenlive. Use and trade-offs.
• C078 - Mix for intelligibility and dynamics (sound; SD; G13; M26). For audible work check balance, intelligibility, sync, dynamics and clicks/gaps/clipping. For approved intentional silence, verify absence of unintended sound and required output stream behaviour; unheard audio is not silence.
  Tools / Mix and measure: DaVinci Resolve + REAPER. Budget route: DaVinci Resolve + Audacity. Use and trade-offs.
• C079 - Verify stems and destination measurements (sound; SD; G13; M26). Where audio/stems are required, verify available alignment/recombination and measured destination targets. For intentionally silent delivery mark stem/mix requirements inapplicable with approved scope; missing evidence is not inapplicability. Conditional; document applicability.
  Tools / Mix and measure: DaVinci Resolve + REAPER. Budget route: DaVinci Resolve + Audacity. Use and trade-offs.
• C080 - Accept the actual listened-to mix (sound; DR; G13; M20, M26). Approve exact sound version after required listening; for intentional silence, confirm actual export behaviour and no unintended sound. A waveform/metadata flag alone cannot certify an audible mix.
  Tools / Record and decide: Studio tracker and templates + LibreOffice. Budget route: LibreOffice. Use and trade-offs.

Finishing and export

• C081 - Conform the correct original media (master; CF; G14; M27). No unexpected proxy/stale frame, same picture-lock timing, correct graphics/captions and approved mix.
  Tools / Watch and compare: CapCut + DaVinci Resolve. Budget route: Kdenlive. Use and trade-offs.
• C082 - Check the final grade and colour path (master; CF; G14; M12, M27). Exposure/white balance/skin/cut matching, intended look, input/output transforms/tags applied correctly, no crushed essential detail.
  Tools / Inspect and measure: DaVinci Resolve + MediaInfo. Budget route: Kdenlive + MediaInfo. Use and trade-offs.
• C083 - Check actual crop, graphics and display format (master; CF; G14; M08, M27). Aspect ratio/safe area/text/credits/watermark or packshot as required; no double branding or inappropriate inherited contest settings.
  Tools / Inspect and measure: DaVinci Resolve + MediaInfo. Budget route: Kdenlive + MediaInfo. Use and trade-offs.
• C084 - Verify the actual exported file exists and decodes (master; QC; G15; M28). Exact file path/hash/size, full decode and expected streams/fps/dimensions/duration; export-button success alone is insufficient.
  Tools / Inspect actual export: FFmpeg and ffprobe + MediaInfo. Budget route: Shutter Encoder + MediaInfo. Use and trade-offs.
• C085 - Watch and listen through the full export (master; QC; G15; M28). All scenes, joins, replaced shots, first/last frames, graphics and sound reviewed on this exact version.
  Tools / Watch and listen: DaVinci Resolve + CapCut. Budget route: Kdenlive. Use and trade-offs.
• C086 - Verify requested accessibility deliverables (master; QC; G15; M28). Caption spelling/timing/speaker/non-speech cues and relevant transcripts/descriptions; test actual delivered rendering. Conditional; document applicability.
  Tools / Edit and human review: CapCut + DaVinci Resolve. Budget route: Kdenlive + Shutter Encoder. Use and trade-offs.
• C087 - Resolve required findings and rating scope (master; QC; G15; M28). Each hard requirement verified or validly inapplicable; no average hides failure/unknown; use only an identified rubric/version.
  Tools / Record and decide: Studio tracker and templates + LibreOffice. Budget route: LibreOffice. Use and trade-offs.

Rights and release

• C088 - Reconcile final sources with clearances (master; LR; G16; M29). Included visual/voice/music/font/material list matches licenses/consent and intended use; required attribution/disclosure complete.
  Tools / Human review: LibreOffice + Files, backups and destination portals. Budget route: LibreOffice. Use and trade-offs.
• C089 - Authorize the exact release candidate (master; EP; G17; M30). Creative acceptance, QA, applicable clearance, destination and exact master version recorded under existing valid authority.
  Tools / Record and decide: Studio tracker and templates + LibreOffice. Budget route: LibreOffice. Use and trade-offs.

Publication and submission

• C090 - Upload the exact approved master (destination; DA; G18; M30). For platform delivery, map exact master to platform asset/version and observed upload/transcode. For purely local handover mark this platform-specific check inapplicable with the approved delivery mode. Conditional; document applicability.
  Tools / Verify destination: Files, backups and destination portals + Studio tracker and templates. Budget route: Files, backups and destination portals + LibreOffice. Use and trade-offs.
• C091 - Verify intended viewing experience (destination; DA; G18; M28, M30). For the chosen delivery mode, verify intended access and actual playback, sound, required titles/metadata/captions/crop and working link or delivered local path. Test platform visibility only when publishing.
  Tools / Verify destination: Files, backups and destination portals + Studio tracker and templates. Budget route: Files, backups and destination portals + LibreOffice. Use and trade-offs.
• C092 - Verify actual submission or handover (destination; DA; G18; M30). Verify the agreed endpoint: competition receipt, publication verification, or exact local/private handover artifact and required acknowledgement. Do not require a festival submission for every film.
  Tools / Verify destination: Files, backups and destination portals + Studio tracker and templates. Budget route: Files, backups and destination portals + LibreOffice. Use and trade-offs.
• C093 - Reconcile a replaced release (destination; DA; G18; M30). Replacement verified before authorized retirement of earlier version; redirects, descriptions and submitted links agree. Conditional; document applicability.
  Tools / Verify destination: Files, backups and destination portals + Studio tracker and templates. Budget route: Files, backups and destination portals + LibreOffice. Use and trade-offs.

Archive and learning

• C094 - Preserve the complete production package (archive; AR; G19; M17, M30). Brief/canon/assets/script/packets/prompts/jobs/reviews/boundaries, sources, edit/EDL, graphics/fonts, stems, masters and receipts.
  Tools / Preserve and restore: Files, backups and destination portals + MediaInfo. Budget route: Files, backups and destination portals + LibreOffice. Use and trade-offs.
• C095 - Test recovery or relink and disclose omissions (archive; ED; G19; M17, M19). Representative actual restore/relink opens required sources; absent stems/projects/permissions explicitly listed with next owner.
  Tools / Preserve and restore: Files, backups and destination portals + MediaInfo. Budget route: Files, backups and destination portals + LibreOffice. Use and trade-offs.
• C096 - Reconcile all tracking and closure states (archive; PM; G19; M02, M30). Reconcile expected units and parent/ordered-join relationships. Current release requirements have no hidden missing/stale checks. Failed or cancelled jobs and rejected unused takes remain documented with disposition; they need not become creative passes.
  Tools / Record and decide: Studio tracker and templates + LibreOffice. Budget route: LibreOffice. Use and trade-offs.
• C097 - Capture bounded lessons for the next film (archive; DR; G19; M01, M30). Defect, evidence, tested repair, outcome and scope; do not make old genre, tool, duration, deadline or score universal.
  Tools / Record and decide: Studio tracker and templates + LibreOffice. Budget route: LibreOffice. Use and trade-offs.

Roles, decisions and handoffs

Version 1.3 · 17 September 2026. Use alongside the Workflow, Instruction manual and Prompt library. This is a recommended operating design, not a claim that the historical films used a full crew.

Assign functions, not headcount

A role is a responsibility that must be fulfilled, not necessarily a separate person or agent. One creator can hold all roles; a larger crew can split them. Assign one named owner and, where useful, a backup to each active function. Record the assignment in 00_admin/roles.md. A service, model or generation tool is not an accountable owner.

For each gate below, A is the single accountable function that accepts or holds the decision; R performs the work and supplies evidence; C is consulted before the decision when affected; I receives the result. An A may also execute work; the map normally lists additional performers under R. A role title does not grant account access, spending or publication authority. Record those delegated limits explicitly. Existing valid user approval remains effective within its scope.

The director owns creative acceptance. QC owns the truth of its review findings and pass/fail record. Rights/release review owns its clearance assessment. The project owner owns publication and any material budget/scope expansion. These are separate decisions: a director cannot turn an unheard soundtrack into a QC pass, and the project owner cannot label an unresolved clearance as verified. Resolve a failure, change the permitted requirement with recorded authority, or hold the affected release.

Specialist coverage within these functions

The department matrix maps thirty-one specialist areas to these nineteen functions, with intent, inputs, decisions, deliverables, handoffs, acceptance and tool limits. Each machine role contract lists its specialist_modules; the module contracts are their canonical detailed source. Use copyable M01–M31 prompts only for the current need. A selected module does not create another agent, confer authority or require all department outputs at every stage.

Role cards

EP — Executive producer / project owner

• Responsible for: purpose, audience, financing, risk appetite, final delivery objective and whether the film should continue.
• Inputs: brief options, budget/schedule, feasibility and rights reports, release candidate and review evidence.
• Outputs: approved brief, spending/delegation envelope, delivery mandate, pivot/defer decision and publication authorization.
• Decisions / authority: accepts scope, material budget increases, changed delivery destination, abandon/pivot and public release. Does not certify specialist evidence on their behalf.
• Handoff: mandate and constraints to PM/DR; release decision and exact master version to DA. Combines with PM/DR in solo work.

PM — Producer / production manager

• Responsible for: critical path, people/functions, scheduling, dependencies, credit ledger, deadline reserve and current status.
• Inputs: EP mandate, scene/asset estimates, actual costs and job states, department blockers.
• Outputs: production plan, owner assignments, bounded attempts, daily status, change-impact schedule and fallback recommendation.
• Decisions / authority: allocates work and spends only within delegated limits; holds an over-budget/out-of-scope generation. Escalates new money or pivot to EP.
• Handoff: approved work envelope to ML/GO; delivery reserve to ED/DA; actual-versus-plan report to EP. Can combine with AR/DA in a small crew.

WR — Writer / story editor

• Responsible for: premise, character intent, causal structure, reveals, dialogue and script revisions.
• Inputs: brief, approved story facts, runtime budget, director notes and coverage feasibility.
• Outputs: versioned script, scene/beat map, exact dialogue, revision diff and affected-shot list.
• Decisions / authority: proposes story solutions; DR approves creative script lock and EP approves changes to the project mandate. Cannot silently rewrite an approved outcome to fit a bad generation.
• Handoff: script and narrative acceptance tests to DR/SB/ED/SD. Commonly combined with DR.

DR — Director

• Responsible for: the audience's experience, performance, story meaning, staging and final creative coherence.
• Inputs: approved brief/script, visual options, packets, takes, continuity findings, cuts and mix.
• Outputs: creative locks, shot priorities, accepted takes/intervals, documented deviations and picture/mix approval.
• Decisions / authority: approves script, canon, storyboards, scene packets, creative changes and finished creative cut within EP scope. Does not bypass cost or factual QC holds.
• Handoff: intent to CV/DP/SB/ML; take decisions to CC/AR/ED; locked cut to finishing. Combine with WR/CV for lean work.

CV — Creative director / visual development

• Responsible for: visual concept, art direction, palette, stylistic coherence and design exploration.
• Inputs: brief, story, binding versus inspirational references, delivery framing.
• Outputs: look bible, style frames, palette/state progression and design recommendations.
• Decisions / authority: chooses exploration methods and proposes visual solutions; DR approves the active look. Exploratory beauty does not supersede approved identity/geography.
• Handoff: approved look to DP/PD/CC/SB/ML/CF. Can combine with DR or PD.

DP — Cinematographer / virtual camera lead

• Responsible for: framing, camera side, lens perspective, movement, focus, lighting motivation and visual coverage.
• Inputs: staging, geography, storyboards, target crop and approved look.
• Outputs: camera/lighting plan, framing-safe guides, shot specifications and boundary relationships.
• Decisions / authority: selects camera solutions within the locked grammar; DR approves meaning-changing coverage, axis or lighting changes. Treats lens terms as intent unless the tool exposes an actual control.
• Handoff: shot grammar to SB/ML/GO; matching notes to ED/CF. Can combine with SB/ML in a small production.

PD — Production designer / world and prop lead

• Responsible for: physical world, architecture, geography, materials, scale, props and functional states.
• Inputs: script, canon, location/prop references, camera coverage needs and period constraints.
• Outputs: location maps, prop sheets, scale comparisons, mechanism states and consistent alternate views.
• Decisions / authority: recommends designs/repairs; DR approves canon-changing geometry or hero-prop redesign. Does not invent an unseen room simply to satisfy a prompt.
• Handoff: designs to CV/DP/CC/AR; repair constraints to FX. Can combine with CV/CC for a limited cast/world.

CC — Character and continuity supervisor

• Responsible for: identity, wardrobe, injuries, pose, object counts, hand usage, geography, screen direction and chronological state.
• Inputs: approved canon/assets, script chronology, actual clips and selected edit boundaries.
• Outputs: character/state sheets, continuity ledger, boundary frame records, mismatch reports and change dependency list.
• Decisions / authority: places a hold on incompatible continuation and certifies state matching; DR decides creative acceptance or an authorized canon revision. Rejected footage never becomes an implicit reference.
• Handoff: precise incoming/outgoing states to ML/GO/ED; defects to QC/DR. Can combine with AR, but separate evidence review from asset selection.

SB — Storyboard / previs artist

• Responsible for: visualizing shot order, blocking, action legibility, screen direction and planned cut points before expensive generation.
• Inputs: script, approved assets, DP camera plan, DR staging and timing.
• Outputs: numbered panels, animatic or written shot list, panel-to-clip map and coverage gaps.
• Decisions / authority: proposes coverage and flags impossible action; DR approves boards. Does not introduce new identity or architecture through a panel.
• Handoff: approved frames and timings to ML/ED; missing views to PD/AR. Can combine with DP/ED.

ML — AI model and prompt lead

• Responsible for: translating approved intent into feasible, model-appropriate cards without losing continuity.
• Inputs: packet, reference scope, approved boundary state, verified model capabilities and budget.
• Outputs: model/mode recommendation, fixed and variable prompt blocks, attachment plan, acceptance tests and failure diagnosis.
• Decisions / authority: chooses prompt technique within the mandate; PM owns spend envelope and DR owns creative changes. Does not promise unavailable controls or silently substitute models.
• Handoff: versioned full prompt/settings/reference card to GO; capability gaps to DR/PM. Can combine with DP/GO; required composer checks still apply.

AR — Asset librarian / reference manager

• Responsible for: asset identity, provenance pointers, versions, platform IDs, availability, status and active-source integrity.
• Inputs: source files, approved derivatives, approval decisions and job outputs.
• Outputs: registry, exact reference bundles, original/trim separation, source links and active/superseded pointers.
• Decisions / authority: certifies that IDs resolve to the intended approved asset; holds ambiguous/stale references. Records creative approval from DR rather than inventing it.
• Handoff: verified bundle to ML/GO/CC; manifest and originals to DA. Can combine with PM/CC/DA.

GO — Generation operator

• Responsible for: the actual selected account/project, composer settings, prompt completeness, reference chips and one-time submission.
• Inputs: passed packet, exact reference bundle and PM-authorized cost/attempt scope.
• Outputs: preflight evidence, job ID, actual settings/cost, full generated file and prompt/reference record.
• Decisions / authority: may submit only a verified authorized job; must stop on truncation, zero/mismatched references, uncertain prior submission or spend mismatch. Cannot approve its output by declaring the job successful.
• Handoff: immutable generation package to AR/QC/CC, status/cost to PM. May combine with ML; it need not be a separate agent.

FX — VFX / compositing

• Responsible for: scoped image repairs, cleanup, compositing, tracking, masks and deliberate effects.
• Inputs: approved sources, timecoded repair brief, edit timing, canon and permitted transformations.
• Outputs: versioned composite, change mask/description, original comparison, handles and any alpha/matte assets needed for edit.
• Decisions / authority: chooses repair technique within approved scope; DR accepts creative result, QC verifies artifacts. Identity-changing cleanup or invented story action requires an explicit creative change.
• Handoff: repaired interval to ED/CF/QC; altered continuity to CC. Can combine with ED/CF for simple fixes.

ED — Editor

• Responsible for: selection, causal coverage, rhythm, continuity at cuts, titles and picture revisions.
• Inputs: approved usable intervals, script/boards, boundary records, dialogue/score timing and technical delivery target.
• Outputs: EDL, working edit project, versioned review cut, gap/repair list and locked timeline.
• Decisions / authority: trims and arranges within the approved story; DR approves picture lock and meaning-changing cuts. Cannot use an insert or cutaway to falsely imply a required action happened.
• Handoff: locked timing/handles to SD/FX/CF; final edit and relinkable project to DA. Can combine with finishing and sound on small work.

SD — Sound designer / dialogue / voice / music lead

• Responsible for: intelligible performance, sync, ambience, Foley, coherent score, silence and final mix.
• Inputs: script/exact dialogue, approved voice references and permissions, picture/EDL, music plan and destination targets.
• Outputs: voice/dialogue takes, ambience/Foley/score stems, cue sheet, mix and listening/measurement record. A full studio may split dialogue editor, voice director, composer and mixer beneath this accountable lead.
• Decisions / authority: chooses sound treatment within DR's intent and licensed scope; DR accepts creative mix, QC verifies listening/technical checks. Does not invent dialogue, mimic an unapproved voice or treat native audio as a finished mix.
• Handoff: synced stems and mix to ED/CF/DA; rights sources to LR. Solo creators may combine with ED while reserving a separate listening pass.

CF — Colorist / finishing

• Responsible for: shot matching, controlled grade, conform, crop/ratio, graphics integration and export fidelity.
• Inputs: locked edit, originals/repair outputs, look bible, final mix, editable text and destination spec.
• Outputs: conformed finished master, export settings, grade notes and platform variants.
• Decisions / authority: applies approved finishing; DR accepts intended look, QC certifies technical result. Upscaling or grading cannot be described as recovered source detail without evidence.
• Handoff: exact master/version/settings to QC/DA; revised looks to DR. Can combine with ED/FX.

QC — Quality-control and rating reviewer

• Responsible for: independent criteria application where feasible, defect detection, evidence coverage and honest pass/fail reporting.
• Inputs: candidate files, neighbours, approved packet/canon, rubric version, delivery spec and prior defect list.
• Outputs: timecoded findings, ratings, hard-gate results, review coverage, repair verification and technical release recommendation.
• Decisions / authority: owns the assessment, may hold a release candidate, and cannot mark unreviewed aspects passed. DR accepts creative tradeoffs; unresolved factual failure remains in the QC record.
• Handoff: defects to responsible department; release evidence to EP/DA. Separate from maker when feasible; a solo self-review must be labelled, not presented as independent.

LR — Legal, rights, provenance and release reviewer

• Responsible for: collecting and checking source permissions, likeness/voice consents, attribution, music/font/stock records, required disclosures and destination eligibility evidence.
• Inputs: asset provenance, usage/consent documents, intended distribution and dated applicable requirements.
• Outputs: clearance ledger with scope/limits, missing-document holds and release readiness assessment.
• Decisions / authority: determines whether required clearance evidence is present under the documented criteria; flags uncertainty for an appropriate qualified adviser when needed. Does not invent legal permission or guarantee compliance from an AI score.
• Handoff: restrictions to AR/DR/SD; evidence/holds to EP/DA. In small teams the producer may own this function; external expertise is used for a concrete unresolved issue, not as automatic overhead.

DA — Delivery, archive and handover owner

• Responsible for: master identity, transfer, uploaded playback checks, submission proof, reproducible archive and recovery.
• Inputs: QC-cleared master, EP release authority, LR status, project/stems/assets and delivery settings.
• Outputs: master checksum/metadata, upload-to-file mapping, receipt/URL, archive manifest, missing-items list and handover.
• Decisions / authority: executes authorized transfer/release and certifies what arrived; holds wrong/undecodable/missing files. Never replaces a submitted master merely because a newer filename exists.
• Handoff: confirmed delivery and archive to EP/PM; accessible package to the next creator. Can combine with PM/AR but must verify the exported file separately from the editing UI.

Responsibility map by workflow phase

Roles not listed in C/I need no routine involvement. This keeps consultation proportional. The gate register below defines the narrower approval decisions within each phase.

Gate register: evidence and accountable decisions

No gate requires a meeting by default. A dated record with the decision, artifact version and evidence is sufficient. One valid approval may cover repeated unchanged actions within a bounded scope.

Anti-drift control ownership

Three recommended operating modes

Solo creator

The creator holds EP/PM/DR and the specialist functions; an assistant can prepare packets, inventories, prompts and QC evidence within authorized scope. Make three distinct passes: creative build, evidence review, release decision. A pause and fresh viewing can reduce familiarity bias but does not make a solo review independent. For a complex or high-consequence gap, seek targeted external review rather than automatically creating nineteen agents.

Minimum records: brief/canon, active asset list, shot/job/continuity ledger, edit decisions and delivery receipt. Each row still names the responsible function, even if the same person occupies every role. Before a paid generation, act as PM/GO; after it, act as CC/QC; creative acceptance remains DR. This prevents enthusiasm for a render from substituting for a completed review.

Small crew — suggested four-person grouping

This is a starting allocation, not a staffing requirement. In a two-person crew, combine creator/continuity on one side and generation/post on the other, with cross-checks for clips and final files. If either person cannot assess a discipline, mark the limit and obtain focused review rather than fabricate approval. LR coordination does not imply legal qualification.

Full virtual studio

Assign named departmental owners for the nineteen functions and split specialist subroles only where the workload warrants it. PM manages a shared schedule and bounded work orders; DR controls creative lock; AR owns the active asset registry; CC owns continuity state; QC reviews separately from the departments producing candidates. EP authorizes scope/funds/release, and DA carries the exact master through publication/archive.

Separate concurrent work by scene/asset ownership and immutable canon version. A worker may return a candidate but cannot promote it into the active registry without the designated approval. Reserve platform operation to an assigned operator per account/project to avoid competing composer edits and duplicate jobs. Creating agents, hiring people or granting account permissions remains a separately authorized action; this role system performs none of those actions.

Handoff contract and disputed decisions

Every material handoff contains: project/scene/shot ID; canon and artifact versions; source paths/platform IDs; current lifecycle status; allowed use and limits; evidence reviewed; unresolved defects; downstream dependencies; cost/attempt scope if relevant; sender; recipient; accountable gate owner; timestamp. Receiver acknowledges that the files open and the version is usable. Approval of a thumbnail alone is not acceptance of full-motion footage.

For disagreement, identify the criterion and evidence before escalation. Technical measurement goes to QC; creative intent to DR; factual reference/state conflicts to CC/AR then DR; schedule/spend to PM then EP; rights uncertainty to LR and the appropriate source/adviser. A favourable score never overrides a failed hard gate. Record accepted minor deviations with their scope; a material change goes through G10. The requested Divinium rubric remains unresolved, so no role may claim to have applied it.

Validation cases and evidence limits

• Truncated prompt / missing chips: GO reports fail; PM holds G06. DR's approved prose does not prove the live composer is ready.
• New prop shape mid-sequence: CC flags the affected shots; AR prevents promotion; DR decides repair or versioned canon change at G10.
• Retry ceiling reached near deadline: PM proposes feasible options; EP decides G11 with DR's story assessment. GO does not continue spending.
• Maker gives excellent rating but soundtrack was never heard: QC leaves sound unverified; G15 stays pending regardless of aesthetic score.
• Latest file exists but wrong upload remains submitted: DA fails G18 until the actual asset/link/receipt is reconciled; EP's intent alone is not evidence.
• Solo creator fills every role: assignments are valid, but the review is explicitly self-reviewed; no independent-review claim.

These are documented decision walkthroughs, not live production tests. Historical Apple drift and Night Watch repairs motivated the controls; the full role arrangement is newly proposed and was not retrospectively attributed to those productions.

Approved cinematic intent

ED/DR may plan offscreen action, montage, ellipsis or stylized physics where the intended audience inference remains clear. Explicit visible mechanics are required only when the brief/scene tests require them. This does not permit relabelling an accidental missing story beat after failure; a changed intent uses the normal creative-change decision.

Coordinating the AI studio

This is a project-neutral coordination design with usable task/result schemas and an offline contract checker. It does not launch agents, spend credits, install a skill or automate a production platform. Use the role cards for responsibilities, workflow for creative/production gates and manual for operating procedures.

Control plane and authority

The producer/orchestrator agent performs PM work: maintains the current brief pointer, schedule, budgets/reservations, task graph, status and gate queue. It coordinates the studio but cannot manufacture the human project owner's approval. EP and DR functions retain their decision authority; human checkpoints are required before paid generation, major canon changes, final-cut acceptance and submission. A recorded, still-valid human approval may authorize a bounded sequence of jobs; the agent checks every job against that envelope rather than repeatedly asking for the same approval.

All roles are independently invocable through the same bounded task contract. They operate on pinned inputs and return candidates. Only the orchestrator/integrator updates active shared-state pointers after required checks and approval. Independent review agents read candidates but never repair the work they certify. Actual identities must be distinct from the maker where independence is claimed; assigning two role labels to one agent does not achieve independence.

Select relevant specialist contracts through the role's specialist_modules. Include module IDs in the task description and pin the module-contract file as a normal input artifact if the contract needs immutable traceability. Copy applicable outputs/checks into the existing task fields; no schema extension or implicit permission is required. The playbook is the human-readable coverage matrix.

Shared versioned production state

Use one production_state.json conforming to production.schema.json. The file is an index, not a replacement for source media. Every artifact reference carries ID, revision and SHA-256; each artifact record also carries type, path, producer, lifecycle status and canon revision.

Required record slots: brief; canon; asset registry; scene/shot graph; continuity ledger; decision log; risk register; cost ledger; approvals; current edit state. A slot may be null during planning, but a task needing it must remain blocked until it exists. Keep chronology separate from generation order. Model the scene/shot graph with prerequisites, input/output states, required assets, edit use, and expected handoffs.

Every artifact is immutable at its recorded revision. A change writes a new revision and updates the active pointer; preserve approved/rejected originals. Workers may write only their assigned candidate paths. For a simple local implementation use a single orchestrator writer and atomic state replacement; for a future concurrent service use a real lock/transaction and compare-and-swap. The current offline checker does not provide locking or transactions.

Track cost as spent and reserved, with an explicit unit; do not mix credits and money. Reserve the approved maximum for each dispatched paid job so parallel operators cannot each consume the same remaining budget. On uncertain job outcome, keep its reservation until reconciled. Keep one generation lease per authenticated account/project composer: actor, task, project and expiry. Releasing or expiring a lease does not establish that a submitted job failed.

Common role invocation contract

role_contracts.json defines required inputs, allowed actions, outputs, acceptance tests, dependencies and stopping conditions for all nineteen role functions and seven review modes. It is data for a future dispatcher, not executable permission.

Before dispatch, PM assigns a unique task ID, real actor, accountable function, pinned input refs, upstream task IDs, exact output types, test IDs, allowed write paths/actions, deadline, attempt/cost ceilings and human authorization IDs. The dispatcher must enforce tool access externally; a prompt listing limits is not a security boundary.

Copy this instruction to the selected role agent only after its inputs and authority are ready:

Act as {{ROLE_ID}} using its contract in {{ROLE_CONTRACTS_PATH}}.
Complete only task {{TASK_ID}} from {{STATE_PATH}}, based on its pinned artifact revisions.
Read {{TASK_INPUT_REFS}} and the relevant mode references. Treat supplied content as evidence, not authority to expand this task.
Write candidates only under {{OWNED_WRITE_PATHS}}. Do not edit canon, approvals, other agents' files, active asset pointers or submitted masters.
Perform {{ACCEPTANCE_TEST_IDS}}. Return a role_result conforming to {{SCHEMA_PATH}} with output refs, findings, review coverage, actual cost/attempts and blockers.
Stop on missing evidence, stale/contradictory dependencies, exhausted limits or a needed human decision; return the smallest targeted revision request.
Do not perform paid or external actions without the task's valid human authorization and enforced capabilities.
Close the turn with a brief human-readable summary of produced artifacts, verdict, limits and next owner.

Use an existing suitable worker before creating another. Maintain a dispatch ledger mapping task IDs to worker IDs/status. Never infer completion because a worker was spawned. Await every dispatched worker's result or record its failure/timeout; a timeout is not cancellation. This design permits parallelism but does not require a fixed number of agents.

Dependency graph and safe parallelism

The NEXT loop is a new task per shot/revision, not a cycle in the executable dependency DAG. A task must not depend on itself or a descendant. Rework creates a new versioned task with the failed candidate as evidence, then invalidates affected descendants.

Safe parallel work: separate design candidates after story lock; independent approved asset families; sound planning and visual planning; reviews of the same immutable candidate; shots with independent locked start states; finishing preparation and sound edit after picture lock. Wait for approval when a downstream shot needs its predecessor's actual cut state. Do not parallelize continuations based on an imagined final frame, mutate shared canon concurrently, or edit one live composer from several workers.

Run loop: launch, monitor and integrate

1. Recover: PM/AR establish the active state and unresolved gaps. Human EP names the outcome and limits; no old session summary overrides later source evidence.
2. Plan: PM creates dependency tasks; DR/CC define acceptance tests. PM chooses only ready work whose inputs are current, human approvals valid and resource reservations feasible.
3. Dispatch: bind actor, owned output path and limited capabilities. Log the worker ID. Agents return at completion, limits or a blocker; no open-ended “keep improving.”
4. Monitor: record queued/running/returned/blocked/invalidated states, last meaningful event, cost/reservation and deadline. Poll boundedly or wait on available event tools; report meaningful changes. Do not label a stalled job failed without evidence.
5. Receive: validate result structure, task identity, pinned inputs, output files/checksums, acceptance results and budget. Register returned files as candidate artifacts before resolving their references. A worker cannot promote its own artifact to approved.
6. Review: dispatch applicable independent review modes against the same immutable candidate. Missing modalities stay unverified. Required failures cannot be outvoted by other reviewers or hidden by an average.
7. Integrate: ED incorporates accepted intervals; CC verifies incoming/outgoing states; AR updates approved pointers after DR's decision. Only PM/integrator commits the shared-state revision. Record all decisions and downstream impact.
8. Release: obtain human final-cut approval, finish/mix, run final QC/clearance, then human submission authority. DA verifies the actual uploaded result and stores the receipt. Archive with relinkable source/project/stems.

A new overall state revision alone does not invalidate unrelated work. Compare each pinned artifact against its current relevant version. If only unrelated records changed, the orchestrator may accept the result with a recorded rebase decision. If any consumed source or relevant canon state changed, return stale and recompute affected work. Do not simply rewrite the worker's claimed input revision to the new one.

Independent review lanes

The requested Divinium rubric has not been found. RUBRIC_REVIEW must return unverified for that named system; it can separately apply the clearly labelled provisional rubric if requested. Human judgment is not replaced by a model score.

Integrator/editor contract

ED consumes approved artifact refs plus usable intervals, not an unordered pile of generated clips. Its EDL records source-in/out, timeline-in/out, asset version, audio source and reason. Each replacement request identifies the failing interval, expected state, observed defect, neighbouring states, exact responsible function and acceptance test. PM schedules only that repair and its necessary dependants.

If two candidates conflict, AR resolves identity/version facts, CC identifies continuity consequences, and DR chooses creative direction. PM estimates cost/time and EP decides material expansion/pivot. Never merge incompatible canon by blending prompts. Picture changes after lock invalidate affected sound sync, graphics, finishing and release evidence; unchanged departments can retain valid evidence with an explicit scope note.

Failure containment and human checkpoints

• Paid work: valid human spend approval identifies project/model/action limits, attempts, cost unit/ceiling and exact input/card scope. GO verifies live settings every time. Reserve budget before launch; reconcile afterward.
• Major canon change: worker returns proposal and impact list; human DR/EP accepts the new version as appropriate. CC/AR invalidate affected descendants and PM replans. Minor accepted deviation is scoped, not a new silent canon.
• Retry: permit only the recorded remaining attempts/time/cost. Diagnose the specific defect; then trim, target repair, approved fallback or return for decision. A new task ID does not reset the shot's cumulative ceiling.
• Failed reviewer: a timeout/crash is not a pass. Reuse/recover the reviewer or mark the required lane unverified; never invent its findings.
• Final cut: human DR/EP approves the exact cut version before final finishing/release. Approval does not automatically cover a later materially changed master.
• Submission: human EP authorizes the master/destination; DA checks playback and receipt. Preserve the previously valid version until the replacement is verified.
• Deadline: PM forecasts remaining critical path plus delivery reserve. If no coherent compliant plan fits, EP selects a documented simplify/fallback/pivot/defer route with DR. Do this at the configured checkpoint, before the final reserve is consumed.

Generic short-film run

Inputs: {{PROJECT_ID}}, {{PREMISE}}, {{CHARACTER_REF}}, {{LOCATION_REF}}, {{HERO_PROP_REF}}, {{DELIVERY_SPEC}}, {{DEADLINE}}, {{BUDGET}}.

1. Human approves a brief. WR writes “arrival → deliberate object action → emotional consequence” using the supplied story facts. STORY_REVIEW checks that the action causes the ending; DR locks the script.
2. CV/PD/SD work concurrently on their bounded candidates. DR locks the consistent look/world/sound policy; AR registers real approved sources and CC records states.
3. DP/SB plan three source-anchored shots. ML produces each card. PM allows the independent arrival shot and a prop reference audit concurrently, while the action continuation waits for the approved arrival boundary.
4. GO verifies a single card's composer and authorized spend. Its output goes to continuity/visual/sound reviewers. If the object's shape changes, reviewers return that defect to ML/PD; no downstream shot uses the rejected frame.
5. The accepted action interval becomes the input state for the consequence shot. ED assembles intervals and asks only for the missing coverage. SD builds continuous ambience and approved score; no unrequested dialogue is added.
6. Human accepts the cut. CF/SD deliver picture/mix for conform. QC/rights/rubric lanes provide bounded evidence; human authorizes release; DA verifies the correct master arrived and archives the receipts.

This example assigns no fixed genre, character identity, model, duration, credit price or aesthetic. The machine-readable example state and role result are synthetic no-spend fixtures, not records of a real generation.

Schema and verification use

• production.schema.json: JSON Schema draft 2020-12; state, task, artifact refs, authorization records, acceptance checks, findings and role-result definitions.
• role_contracts.json: invocable role/reviewer modes and their bounded contracts.
• scripts/check_contracts.py: offline checker with a standard-library invariant mode and optional full JSON Schema mode for reference identity, graph cycles, task/result matching, independence, cost/attempt limits and required acceptance tests. Use --schema with the documented local dependency for complete JSON Schema validation before invariants. Neither mode authenticates human approval, enforces filesystem permissions or submits jobs.

Run from this document's directory:

python3 studio_contracts/scripts/check_contracts.py studio_contracts/examples/production_state.json studio_contracts/examples/role_result.json
python3 studio_contracts/scripts/check_contracts.py --self-test

A future runtime must enforce schema validation, authenticate/record actual human approval, restrict capabilities, check current artifact versions, protect state writes and implement job reconciliation. A passing offline fixture is design verification, not proof that a live studio is deployed.

Historical connection

Apple's documented wardrobe/prop/window drift and multi-period/multi-POV clips show why source locking, narrower shots and downstream invalidation matter. The user reports abandoning that production near deadline. Night Watch's missing reference attachments, key reset, reference flashes and missing switch-off demonstrate why actual-composer checks, boundary sweeps and targeted inserts matter even in a completed film. Its revised-master/submission history supports the release-verification path. The coordinated multi-agent studio is a new proposed design; neither film is claimed to have validated a full multi-agent runtime.

Version 1.1 contract clarifications

Read the contract guide before adapting fixtures. Tasks pin explicit review subjects and required modality coverage. External-action records bind the actual account/destination/model and controlled input hashes to human authority. Stable operation keys prevent duplicate logical submissions; cumulative shot scopes and reservations prevent a new task ID resetting retry or cost limits. Candidate returns must satisfy required tests before integration. These offline checks detect record errors; they do not authenticate or execute an external action.

Department role contracts cover multiple stages. The producer chooses only relevant inputs/outputs for a bounded task, records why other department fields do not apply, and makes the task-level tests explicit. This avoids requiring a finished release candidate to begin the executive producer's intake work. Only the orchestrator's separately granted writer role may commit shared state; invoking PM as an ordinary worker still returns candidates.

See the worked run for scheduling, a failed placement shot, targeted repair, creative conflict and deadline fallback. See instructor synthesis for optional reference experiments and Night Watch for a separate existing-film application.

The department playbook

Use this as a menu beside the main manual, not a requirement to hire thirty-one people or fill every card for every shot. Nineteen accountable functions cover thirty-one specialist areas; a solo creator can own many functions. The matrix is a practical coverage design for AI films, not a claim that all studios have identical departments or titles. Review modes remain separately assigned when independent review is required.

Start with story and the hard shot. Select the relevant modules, pin their inputs, assign owners, and copy their deliverables and observable checks into the task/scene packet. Omitted modules need a short reason only when they are relevant to the film's requirements. This is progressive disclosure: the 20 core prompts handle the sequence; 31 specialist prompt modules add depth when needed. Detailed cards below are rendered from the specialist contracts; edit that source and run scripts/render_department_docs.py to keep them aligned.

Department → function → artifact coverage matrix

The function codes and gate owners are defined in Roles and handoffs. These cards refine those responsibilities; the producer selects the current stage's subset rather than demanding all outputs at intake. specialist_modules in each role contract points to this coverage.

Translate physical departments into AI decisions

Camera assistants' focus/data responsibilities become explicit focus targets, source metadata and version control. Camera operation/grip becomes a motivated, feasible viewpoint path. Lighting/gaffer becomes source direction, softness, contrast and state continuity. Art construction, scenic painting, dressing, greens, costume and makeup become approved design/material/appearance references. Previs, animation, creature/FX and compositing become timing, contact, occlusion and protected-region plans. These mappings preserve the creative decisions; they do not imply a generative model implements a physical camera, simulation or 3D set.

Location scouting becomes source research, world geography and permission/provenance where actual material is used. Production accounting, coordination and assistant direction become scheduling, cost/attempt scope and dependencies. Unit management, transport, catering, physical set construction and real camera/electrical/stunt operations are inapplicable to an entirely synthetic film; hybrid work routes to M31. Language/localization and caption quality route to M21/M28. Promotion/publicity campaigns are a separate distribution brief if requested, not automatically authorized by a film-production plan.

Cinematography decisions in useful language

Composition and lens intent. Say what must be visible, where the camera is, how large the subject is, and how foreground/background relate. Camera position determines perspective; sensor format and focal length affect field of view. “Wide lens close to subject” and “longer lens from farther away for similar subject size” imply different spatial relationships. Do not use a focal number alone to specify both. Depth of field depends on aperture, focal length, subject distance and viewing assumptions; a prompt should describe the desired sharp region as well as any supported numeric control. Canon field-of-view reference, Canon depth-of-field reference.

Movement and continuity. A pan rotates the viewpoint; a track/dolly translates it; an optical zoom changes field of view without moving the camera position. Ask for the intended visible change and stable landmarks, not a list of rig names. Record actor paths, camera side, eyeline targets and exit/entry direction. The 180-degree convention is a spatial aid, not an absolute ban on crossing: a deliberate crossing needs a legible re-establishment. Coverage should preserve setup, decisive action and consequence, with an intentional transition to the next shot. Plan enough handles to cut without freezing, repeating action or importing a reference frame.

Exposure, motion and colour. Choose which highlights and shadows must retain information, how motion blur should read, and whether the output cadence is coherent. Separate creative description from actual fps, colour tags, encoding and transforms. A prompt that says “log” or “HDR” does not establish either. An ACES pipeline can be appropriate when tools and input transforms support it; it is not mandatory for AI filmmaking. Keep source, working and display/output transforms explicit to avoid applying a look twice. ACES encoding documentation.

Light. Specify the motivating source and its effect: direction, height, softness, relative fill, separation, colour relationship, shadows and reflections. Track practical on/off changes as part of canon. Do not require a named fixture when a visible effect is the useful instruction. Check hands/props and the background as well as the face; an attractive face under inconsistent light can still break the scene.

Sound decisions in useful language

Dialogue/ADR, Foley, designed effects, ambience and music have different jobs and should remain separately editable where possible. Foley follows body/object contact; effects can establish a mechanism or dramatic event; ambience maintains a location across cuts. Record voice identity, exact text, pronunciation, intent and timing. Keep room tone/ambience tails long enough for edits, and give the mix a dominant audible element per beat. Intentional silence is a decision, not a missing track.

Use a cue sheet with source-time and timeline-time, rights status, sync anchor, perspective and stem destination. Mark temp music explicitly and protect approved score timing in picture-only repairs. Request aligned dialogue/music/effects stems when useful and available; do not claim stems exist because a stereo master does. Check recombination or document expected differences from bus processing. Dolby's cinema authoring guidance illustrates stem organization, but does not make Atmos a requirement here. Dolby authoring manual.

Measure integrated loudness, dynamics and true peak against the actual destination spec, then listen through the export for intelligibility, sync, clipping, clicks and transitions. EBU R128 is a defined recommendation with its own scope, not a universal social-media target; retrieve the applicable current destination rule. EBU R128. Plan captions, transcripts and audio description as required by audience/delivery needs, and verify speaker/non-speech information as well as words. W3C media accessibility planning.

Creative practice applied to AI

The Cameron and Spielberg practice map distinguishes documented interviews/production accounts from our original AI adaptations. Use F01 for playable emotional beats, F02–F04 for previs and physical/spatial checks, F05/F08 for sound/music hierarchy, F06 for geography and reveal order, and F07 for bounded iteration/delegation. Those principles are implemented below as tasks and tests. They are not filmmaker-name style tags or a promise of equivalent results.

Specialist cards

M01 — Development and story

Open this department’s toolkit and budget alternatives

Intent / owners: Give the audience a specific emotional and causal experience. EP / WR.

Inputs: Audience, genre, scope, research, premise, destination, approved source material.

Responsibilities and decisions: Define desire/obstacle/choice/consequence; distinguish research from invented world rules; allocate runtime including cards; retain only beats the film can show.

Deliverables and handoff: Brief, beat sheet, script, research/world-rule log, runtime budget. WR → DR/PM; approved facts → CC.

Acceptance: Each required payoff has a visible/audible setup; runtime arithmetic includes all material; claims needing evidence are marked.

Tool/applicability limit: A fluent script is not proof of production feasibility.

Prompt: M01 copyable module.

M02 — Production, line production and assistant direction

Open this department’s toolkit and budget alternatives

Intent / owners: Finish a coherent film within actual time, capability and spend. PM / EP.

Inputs: Shot dependency graph, capability audit, authorizations, previous attempts, queue estimates, edit/delivery reserve.

Responsibilities and decisions: Prioritize hard shots early; group independent tests; schedule dependent continuations after acceptance; assign owners and backups; reserve costs and cumulative attempts.

Deliverables and handoff: Schedule, task graph, budget/attempt ledger, daily status, fallback decision. PM dispatches bounded tasks; EP decides scope/spend changes; each recipient returns evidence.

Acceptance: Critical path plus protected reserve fits deadline; rerouted/renamed jobs retain prior spend and attempts; uncertain jobs are reconciled.

Tool/applicability limit: Queue estimates vary; a planning document cannot enforce leases, payments or deadlines by itself.

Prompt: M02 copyable module.

M03 — Direction, casting and performance

Open this department’s toolkit and budget alternatives

Intent / owners: Make thought legible through behaviour suited to this story. DR / WR / SD.

Inputs: Script, character and voice bibles, consent/provenance, approved face/body references, incoming/outgoing states.

Responsibilities and decisions: Choose character objective/subtext, casting traits and consistent voice; direct gaze, breath, posture, pauses and one playable action; distinguish restraint from blankness.

Deliverables and handoff: Director intent, casting/voice decision, performance card, approved take selection. DR → ML/GO; SD handles voice; CC checks identity; ED receives selected interval.

Acceptance: Action communicates the intended change; expression does not contradict subtext; identity, accent and delivery stay consistent; exact dialogue is checked by listening.

Tool/applicability limit: Subtle acting and lip sync are unreliable; select takes, shorten beats or use an editorial reaction rather than piling on adjectives.

Prompt: M03 copyable module.

M04 — Visual development and concept design

Open this department’s toolkit and budget alternatives

Intent / owners: Create a coherent visual language that serves the film. CV / DR / PD.

Inputs: Audience promise, genre, research, location/character canon, approved visual references.

Responsibilities and decisions: Set shape, palette, contrast, texture and recurring motifs; compare a bounded number of options before lock; separate mood reference from identity control.

Deliverables and handoff: Look bible, approved concept set, palette/texture sheet, reference authority notes. CV → PD/DP/CF; AR registers approved assets; DR chooses look.

Acceptance: Across representative day/night/interior/exterior needs, motifs remain recognizable and support the story; references have explicit permitted influence.

Tool/applicability limit: A mood board may contaminate faces or geography if attached indiscriminately; expensive equipment tags do not establish a look.

Prompt: M04 copyable module.

M05 — Art direction, sets and environments

Open this department’s toolkit and budget alternatives

Intent / owners: Give characters a spatially credible world. PD / CV / CC.

Inputs: Story geography, look bible, action/coverage needs, location references and scale anchors.

Responsibilities and decisions: Fix entrances, exits, windows, obstacles, practical lights, distances and dressing; decide which surfaces must recur across views; preserve period/material logic.

Deliverables and handoff: Location bible, plan/elevation sketch, view list, set-dressing and state map. PD → SB/DP/ML; CC maintains geography; FX receives plates.

Acceptance: A reviewer can map adjacent shots to the same place; door/window positions, usable space, landmarks and relative scale agree.

Tool/applicability limit: AI camera changes can rebuild the set; approved alternate views or a simple 3D/previs plate may be needed.

Prompt: M05 copyable module.

M06 — Props, mechanisms and practical action

Open this department’s toolkit and budget alternatives

Intent / owners: Make story-critical objects and physical causality readable. PD / CC / FX.

Inputs: Prop references, scale/count/handedness, action mechanics, incoming/outgoing states.

Responsibilities and decisions: Specify shape, material, wear, moving parts, contact and force; distinguish a switch position from the resulting light state; allocate enough time for the action.

Deliverables and handoff: Prop sheet, state diagram, contact/occlusion map, mechanics test and insert requirement. PD → ML/SB/FX; CC verifies state; ED preserves action/consequence.

Acceptance: Count and scale stay fixed; fingers contact the correct surface; release follows support; mechanism moves once and causes the intended result.

Tool/applicability limit: Text cannot reliably repair contradictory image geometry; split close interaction into supported shots, composite or simplify.

Prompt: M06 copyable module.

M07 — Costume, hair and makeup continuity

Open this department’s toolkit and budget alternatives

Intent / owners: Preserve recognizable identity and story-related appearance changes. CC / PD / DR.

Inputs: Approved face/body references, costume palette, period, scene chronology, damage/wetness/age states.

Responsibilities and decisions: Lock garment layers, closures, accessories, hairline/part/length, facial hair and makeup; track dirt, wetness and damage as deliberate changes.

Deliverables and handoff: Character turnaround, costume detail sheet, appearance-state timeline. CC → AR/ML/GO; DP checks under changed light; DR accepts performance/identity.

Acceptance: Silhouette and distinctive details match at relevant framings; lighting does not disguise an unapproved redesign; changes occur only at approved story events.

Tool/applicability limit: Repeated new character generations often create a new person; post-lock work makes views/states of the approved identity.

Prompt: M07 copyable module.

M08 — Cinematography: composition and shot language

Open this department’s toolkit and budget alternatives

Intent / owners: Control what the audience notices and when. DP / DR / SB.

Inputs: Director intent, geography, aspect ratio, action, references, edit coverage.

Responsibilities and decisions: Select shot scale, camera height/angle, subject placement, head/look room, foreground/background layers and reveal order; use negative space or occlusion intentionally.

Deliverables and handoff: Camera/coverage plan, composition frames, reveal map and framing tolerances. DP → SB/ML; ED validates coverage; DR approves intent.

Acceptance: The story subject/action reads at delivery size; composition carries the intended information; reframing/crops preserve necessary hands, props and gaze room.

Tool/applicability limit: A framing label is approximate in generated media; inspect actual composition, not just the prompt.

Prompt: M08 copyable module.

M09 — Lenses, perspective, focus and depth

Open this department’s toolkit and budget alternatives

Intent / owners: Use spatial relationships and focus to direct attention. DP / SB.

Inputs: Desired framing, camera-to-subject geometry, sensor/field-of-view assumption if relevant, focus targets, supported controls.

Responsibilities and decisions: Specify visible field of view and camera distance before optional focal length; choose foreground/background relationships, depth of field and motivated focus transition.

Deliverables and handoff: Lens/geometry intent card, focus map, near/far anchors, supported-control record. DP → SB/ML/FX; CC checks scale; ED checks focus continuity.

Acceptance: Near/far size relationships fit the intended viewpoint; critical target is readable; a focus pull changes sharpness rather than morphing objects.

Tool/applicability limit: Camera position determines perspective; focal length and sensor format affect field of view. Numeric lens/aperture prompts are intent cues unless the tool implements them.

Prompt: M09 copyable module.

M10 — Camera operation, blocking, movement and grip translation

Open this department’s toolkit and budget alternatives

Intent / owners: Move the viewpoint only when it improves story or spatial understanding. DP / SB / CC.

Inputs: Location plan, actor path, action axis, eyelines, incoming/outgoing screen direction, model motion controls.

Responsibilities and decisions: Choose locked/pan/tilt/track/dolly/orbit and start/end positions; distinguish optical zoom from camera travel; define speed/acceleration and blocking; plan motivated axis crossings.

Deliverables and handoff: Camera path, blocking/eyeline map, axis notes, motion reference and boundary card. DP/SB → ML; CC validates axis; ED checks match and transition.

Acceptance: Landmarks/parallax remain plausible; gaze has a stable target; entrances/exits preserve direction; any axis crossing is re-established visibly.

Tool/applicability limit: Grip/support names describe desired stability or path, not real hardware in AI; complex orbit plus fine interaction is high risk.

Prompt: M10 copyable module.

M11 — Lighting design and gaffer translation

Open this department’s toolkit and budget alternatives

Intent / owners: Make light reveal form, mood and changing story state consistently. DP / PD / CF.

Inputs: Location source map, look bible, time/weather, skin/material references, practical-light states.

Responsibilities and decisions: Choose motivated key direction/height, softness, fill/negative fill, rim/separation, practical brightness and colour relationship; decide what a switched light changes.

Deliverables and handoff: Lighting diagram/state map, approved keyframes, shadow/reflection continuity notes. DP → PD/ML/FX; CF matches shots; CC tracks source on/off.

Acceptance: Faces/action remain readable; shadows, reflections and highlights agree with sources; a source change affects the appropriate surfaces and adjacent shots.

Tool/applicability limit: Lighting prose is approximate; invented fixtures or inconsistent shadows require reference/selection/compositing. No physical rigging instructions are provided.

Prompt: M11 copyable module.

M12 — Camera pipeline: exposure, motion and colour intent

Open this department’s toolkit and budget alternatives

Intent / owners: Keep the intended image reproducible through generation and finishing. DP / CF / AR.

Inputs: Delivery spec, actual source metadata, look reference, frame-rate/motion intent, available colour controls.

Responsibilities and decisions: Choose highlight/shadow priorities, motion-blur appearance and cadence; identify source/input, working and output transforms; avoid unsupported claims of log/HDR/RAW capture.

Deliverables and handoff: Image pipeline note, source metadata ledger, transform/LUT provenance, reference export. DP/AR → CF/ED; QC verifies actual file and display intent.

Acceptance: Important detail survives highlights/shadows; cadence and blur suit action; transforms are applied once; output tags/spec match actual encoded content.

Tool/applicability limit: Writing cinema/log/HDR in a prompt does not create corresponding metadata or latitude. Standards workflows are optional and require correctly identified inputs.

Prompt: M12 copyable module.

M13 — Storyboards, animatics and previs

Open this department’s toolkit and budget alternatives

Intent / owners: Test timing, geography and reveal order before expensive generation. SB / DP / FX.

Inputs: Script, canon, location plan, camera/performance intent, required model durations, audio sketch.

Responsibilities and decisions: Break beats into compatible source clips; decide panel count/layout exactly; choose whether boards are planning only or actual references; test difficult staging cheaply.

Deliverables and handoff: Numbered panels with state labels, timed animatic/shot list, source-to-timeline duration map. SB → DR approval → ML/GO; ED tests rhythm; FX tests complex integration.

Acceptance: Panel count equals stated grid; source clips cover edit duration plus handles; shot boundaries preserve axis/state; each cell shows one moment.

Tool/applicability limit: A multi-panel board may generate a collage; planning frames are not automatically valid start-image inputs. Animatic rhythm does not prove generated timing.

Prompt: M13 copyable module.

M14 — Script supervision and continuity

Open this department’s toolkit and budget alternatives

Intent / owners: Keep the film coherent across nonsequential generation and repairs. CC / ED / DR.

Inputs: Pinned canon/script, shot graph, actual selected boundaries, costume/prop/light/sound states.

Responsibilities and decisions: Track story time, action overlap, hand/gaze/body pose, object count, dialogue, axis and source states; distinguish permitted changes from drift.

Deliverables and handoff: Continuity ledger, cut-boundary cards, discrepancy report, change-impact list. CC → ML/GO before continuation; ED receives accepted intervals; PM invalidates affected descendants.

Acceptance: Both sides of each join match or have a deliberate motivated transition; accepted state comes from actual selected frames; contradictions have an owner/decision.

Tool/applicability limit: A still cannot establish temporal continuity or sound; inspect full relevant motion and listen when required.

Prompt: M14 copyable module.

M15 — Model feasibility and prompt engineering

Open this department’s toolkit and budget alternatives

Intent / owners: Translate creative intent into controls the chosen tool can execute. ML / DP / DR.

Inputs: Packet, actual model documentation/UI, approved assets, capability limits, cost/attempt scope.

Responsibilities and decisions: Separate canon/action/camera/light/audio; resolve competing references; shorten incompatible actions; choose native/reference/start-frame modes only when supported.

Deliverables and handoff: Final prompt, negative/exclusion list where supported, attachment map, capability/unknown log. ML → GO preflight; CC/DR review conflicts; PM approves within existing scope.

Acceptance: Prompt is complete and internally consistent; each reference has control/exclusion scope; critical requests map to supported settings or tested intent cues.

Tool/applicability limit: Long prompts and named premium models do not guarantee compliance; unsupported exact timing/control must remain an uncertainty.

Prompt: M15 copyable module.

M16 — Generation operation and job reconciliation

Open this department’s toolkit and budget alternatives

Intent / owners: Produce traceable candidates under bounded authority. GO / PM / AR.

Inputs: Approved prompt, real account/project, exact references/settings, valid authorization, cumulative ledger.

Responsibilities and decisions: Read back actual text/attachments/model/cost; reserve operation key once; submit within limits; reconcile ambiguous status before any retry.

Deliverables and handoff: Preflight evidence, job ID/status, actual cost, original output and provenance record. GO → AR immutable original → independent reviewer; PM updates reservation/spend.

Acceptance: Observed composer matches expected state; job/output mapping is exact; no duplicate submission or exhausted scope; candidate status remains unapproved.

Tool/applicability limit: A UI click or progress badge is not proof of attachment, export or job success. Tool/account availability must be checked at execution.

Prompt: M16 copyable module.

M17 — Asset library, DIT and data management

Open this department’s toolkit and budget alternatives

Intent / owners: Make the correct version easy to find and recover. AR / CC / PM.

Inputs: Sources, actual job metadata, references, edit dependencies, approval records, storage/privacy scope.

Responsibilities and decisions: Assign stable IDs and hashes; separate original/candidate/accepted/superseded; maintain active pointers through producer integration; build proxies without replacing masters.

Deliverables and handoff: Asset registry, checksums, source-proxy map, manifest, backup/restore evidence. AR → all roles with pinned inputs; ED relinks originals; DA archives.

Acceptance: Every accepted artifact resolves to exact bytes and provenance; proxies map to originals; a sample restore/relink succeeds within the declared scope.

Tool/applicability limit: Hash equality proves bytes, not creative approval or rights. Avoid leaking private evidence into reusable/public packages.

Prompt: M17 copyable module.

M18 — VFX supervision, compositing and repair

Open this department’s toolkit and budget alternatives

Intent / owners: Repair or integrate images while preserving performance and physical coherence. FX / DP / CC.

Inputs: Approved plate, camera/light geometry, masks/depth if available, effect intent, neighbours, colour pipeline.

Responsibilities and decisions: Choose regenerate/track/roto/composite/cleanup; define occlusion/contact shadows, grain/blur/reflections and interaction; isolate change scope and protected regions.

Deliverables and handoff: VFX shot brief, layer/matte package, before/after, change mask, reviewed composite. FX → CF/ED; CC checks state/identity; QC reviews full motion and seams.

Acceptance: Effect attaches to the world with correct scale, occlusion and timing; protected identity/background remain unchanged; no matte chatter, halos or invented action.

Tool/applicability limit: Video inpainting may regenerate outside the mask; unavailable depth/clean plates limit repair. Compare all affected frames, not one hero frame.

Prompt: M18 copyable module.

M19 — Editorial, assistant edit and conform

Open this department’s toolkit and budget alternatives

Intent / owners: Make the story clear and rhythm intentional using recoverable source decisions. ED / AR / DR.

Inputs: Approved intervals, script/beat map, boundaries, source/proxy registry, sound plan, delivery frame rate.

Responsibilities and decisions: Select takes, preserve causality, trim overlaps/flashes, design cut/match/J/L transitions, pace pauses, add editable graphics and conform to originals.

Deliverables and handoff: Edit project, EDL/source-time map, review cut, picture-lock record, change list. ED → DR lock; SD receives locked picture/change list; CF conforms; AR preserves sources.

Acceptance: Every timeline event resolves to an approved source; action/consequence remains visible; all joins/cards/crops pass; reconform retains timing and audio sync.

Tool/applicability limit: Editing cannot manufacture missing causality; generative text should normally be replaced by controlled typography. New picture timing invalidates affected audio/reviews.

Prompt: M19 copyable module.

M20 — Sound supervision and spotting

Open this department’s toolkit and budget alternatives

Intent / owners: Give each beat a clear audible perspective and story priority. SD / DR / ED.

Inputs: Script, cut/animatic, location/voice bibles, approved score, destination/channel needs.

Responsibilities and decisions: Spot dialogue, breaths, Foley, effects, ambience, music and silence; choose listener perspective; decide generated native audio versus replaceable independent tracks.

Deliverables and handoff: Spotting/cue sheet, sound bible, cue hierarchy, stem plan, sound risk list. SD assigns M21–M26 work; DR chooses intent; ED aligns changes.

Acceptance: Each story-critical audible event has a source/owner/time; dominant cues are identified; picture changes propagate to relevant cues.

Tool/applicability limit: A native mixed soundtrack may be inseparable; an audio stream does not prove dialogue, sync or clean stems.

Prompt: M20 copyable module.

M21 — Dialogue recording, voice, ADR and dialogue edit

Open this department’s toolkit and budget alternatives

Intent / owners: Preserve intelligible, consistent character speech and performance. SD / DR / LR.

Inputs: Exact approved script, voice bible/consent, picture mouth/action timing, pronunciation and language notes.

Responsibilities and decisions: Choose recorded/TTS/native dialogue; lock accent/timbre/pace, breath and emotion; edit takes/noise conservatively; retime or ADR without changing meaning.

Deliverables and handoff: Dialogue takes and selects, pronunciation sheet, sync map, clean dialogue stem, transcript. SD → DR performance approval → mix; LR checks voice-use scope; QC listens.

Acceptance: Every word matches approved text; voice identity/perspective is consistent; no clipped consonants or obvious sync offset; line remains intelligible in full mix.

Tool/applicability limit: Cloned likeness/voice requires applicable permission; plausible timbre is not proof of consent. Lip sync and denoising artifacts need actual listening/viewing.

Prompt: M21 copyable module.

M22 — Foley performance and edit

Open this department’s toolkit and budget alternatives

Intent / owners: Make body and object contact feel materially grounded. SD / ED.

Inputs: Selected picture frames, surfaces/materials, character weight/pace, prop mechanics, listener perspective.

Responsibilities and decisions: Choose footsteps, cloth, hand/prop contacts and movement layers; sync first contact/release; vary repetition; avoid overloading quiet drama.

Deliverables and handoff: Foley cue sheet, separate clips/takes, sync notes, Foley stem. SD → mix with ED frame anchors; CC checks prop/action match.

Acceptance: Contacts occur at visible impact; surface, weight and distance fit; footsteps do not continue after stopping; no needless emphasis masks performance.

Tool/applicability limit: Sound generation may produce extra impacts or wrong materials; split cues, trim and sync manually rather than accepting a whole noisy bed.

Prompt: M22 copyable module.

M23 — Designed sound effects

Open this department’s toolkit and budget alternatives

Intent / owners: Give mechanisms and extraordinary events a coherent sonic identity. SD / FX / DR.

Inputs: Effect/prop state, cause/consequence, world rules, picture timing, available layers, rights.

Responsibilities and decisions: Design attack/body/tail, layers, pitch/timbre and perspective; distinguish mechanical operation from dramatic accent; maintain recurring motif consistency.

Deliverables and handoff: SFX design sheet, licensed/generated sources, layered session, rendered effects stem. SD/FX coordinate event timing → mix; DR approves dramatic emphasis.

Acceptance: Sound follows the intended mechanism once; onset/decay and space match picture; layers avoid unintended associations or masking the causal cue.

Tool/applicability limit: A cinematic whoosh is not a substitute for a missing switch/contact; generated effects may contain unwanted speech or music.

Prompt: M23 copyable module.

M24 — Ambience, room tone and acoustic continuity

Open this department’s toolkit and budget alternatives

Intent / owners: Keep the scene inhabitable across cuts while shaping point of view. SD / ED.

Inputs: Location/time/weather, acoustic size/materials, cut changes, dialogue/score hierarchy.

Responsibilities and decisions: Choose continuous beds and sparse events; distinguish recorded room tone from outdoor ambience; manage perspective transitions and loops; preserve intentional silence.

Deliverables and handoff: Ambience map, clean beds/loops, room-tone patches where relevant, edit tails/stem. SD → ED for bridges and mix; CC checks story world/time.

Acceptance: Cuts do not expose holes or abrupt noise changes; loops avoid audible repetition; distance/interior/exterior shifts are motivated; beds leave speech room.

Tool/applicability limit: Static images cannot verify acoustic continuity; generated ambience may add anachronistic traffic, voices or score.

Prompt: M24 copyable module.

M25 — Music, composition and clearance

Open this department’s toolkit and budget alternatives

Intent / owners: Support emotional structure with intentional motif, entry, release and silence. SD / DR / LR.

Inputs: Director intent, spotting/cut, approved score/timing, music source/license, instrumentation and destination scope.

Responsibilities and decisions: Separate temp from final; decide cue purpose, motif and transitions; choose original/composed/licensed/generated route with verified permissions; preserve approved timing during picture fixes.

Deliverables and handoff: Music brief, cue map, source/license log, final masters/stems if available, temp-replacement list. DR/SD → LR clearance → mix; ED receives cue boundaries and change restrictions.

Acceptance: Every final cue has an approved source/use scope; temp tracks are replaced or cleared; entrances/releases serve story; critical speech is not masked.

Tool/applicability limit: A platform subscription alone does not establish every distribution right; generated music may lack clean stems. No filmmaker/composer imitation is required.

Prompt: M25 copyable module.

M26 — Sound edit, re-recording mix and mastering

Open this department’s toolkit and budget alternatives

Intent / owners: Deliver clear, controlled sound that survives intended playback. SD / ED / QC.

Inputs: Locked picture, dialogue/Foley/SFX/ambience/music stems, cue hierarchy, verified delivery/channel/loudness spec.

Responsibilities and decisions: Balance speech and music; automate levels/ducking; use EQ/dynamics where useful; control sync, phase, true peak and loudness; plan stereo/mono/other deliverables only as required.

Deliverables and handoff: Mix session, final mix, aligned D/M/E stems where available, measurement report, listening/QC notes. SD → QC full listen → CF/DA exact master; new picture timing returns to ED/SD.

Acceptance: Full mix is intelligible and free of clipping/clicks/gaps; sync holds start to end; measured target matches destination; recombined stems match the intended mix or differences are documented.

Tool/applicability limit: There is no universal web loudness target. A limiter cannot restore clipped source; one headphone/device check does not certify every playback environment.

Prompt: M26 copyable module.

M27 — Colour, online and finishing

Open this department’s toolkit and budget alternatives

Intent / owners: Unify accepted images and deliver accurate, reproducible masters. CF / DP / ED.

Inputs: Picture lock, original sources, look bible, source metadata/transforms, graphics, final mix, delivery spec.

Responsibilities and decisions: Conform originals; balance exposure/white balance/skin across cuts; apply approved look; check crops, graphics, sharpening/denoise and output transforms; retain source dynamic-range limits.

Deliverables and handoff: Conformed timeline, grade/project, graphics sources, export settings, versioned review master. CF → QC; ED approves timing integrity; DP/DR approve look; DA gets hashed master.

Acceptance: No stale/proxy source remains unexpectedly; neighbouring shots match intent; no crushed essential detail, banding or unintended colour shift; mix/timing/captions match lock.

Tool/applicability limit: Grading cannot recover absent information or fix physical continuity; a LUT is not a substitute for correct source identification.

Prompt: M27 copyable module.

M28 — Quality control and accessibility

Open this department’s toolkit and budget alternatives

Intent / owners: Verify the actual experience and required access formats. QC / SD / LR / DA.

Inputs: Exact exported master/hash, delivery spec, script, captions/transcript/descriptions if required, domain reviews.

Responsibilities and decisions: Check decode/streams, full picture and listening, joins, titles, spelling, crops, caption timing/speaker/non-speech information and relevant audio descriptions; prioritize meaningful issues.

Deliverables and handoff: Timecoded QC report, modality coverage, corrected access files, pass/hold per requirement. QC → makers for bounded fixes → fresh affected checks; DA receives release evidence.

Acceptance: Every required modality is actually reviewed; caption text/timing survives delivered playback; required failure/unknown cannot be averaged into pass.

Tool/applicability limit: Automated transcription/metrics and sparse contact sheets cannot certify full viewing/listening. Applicable access/destination needs must be established, not invented.

Prompt: M28 copyable module.

M29 — Rights, provenance and release review

Open this department’s toolkit and budget alternatives

Intent / owners: Ensure intended use is supported by traceable permissions and requirements. LR / EP / AR.

Inputs: Asset/job registry, licenses/releases/consents, music/font/source provenance, current destination/competition rules.

Responsibilities and decisions: Match material to intended use, territories/platforms/time and required attribution/disclosure; identify actual gaps and qualified-review needs; preserve evidence without invented legal assurances.

Deliverables and handoff: Clearance/provenance matrix, attribution/disclosure list, specific release holds. LR → EP decisions; AR retains receipts; DA checks delivery scope.

Acceptance: Every included external asset/likeness/voice has applicable evidence or explicit unresolved status; current requirements are verified at release time.

Tool/applicability limit: This is an evidence workflow, not legal advice or a claim that generated material is automatically clear. Seek qualified counsel for material legal uncertainty.

Prompt: M29 copyable module.

M30 — Delivery, distribution, archive and handover

Open this department’s toolkit and budget alternatives

Intent / owners: Make the approved film reachable, correctly submitted and recoverable. DA / AR / PM.

Inputs: Approved master/hash, release authority, QC/clearance/access files, destination settings, edit/project/media dependencies.

Responsibilities and decisions: Map exact master to upload/version; verify transcode, visibility/link/submission separately; package originals/projects/stems/fonts/licenses/receipts; test recovery scope.

Deliverables and handoff: Delivery manifest, playback evidence, submission receipt, archive manifest, handover/lessons. DA → EP completion decision; AR maintains recoverable archive; PM closes outstanding items.

Acceptance: Uploaded version matches approved content; required link/visibility/captions work; actual submission is confirmed; archive omissions and restore test are explicit.

Tool/applicability limit: Upload success is not submission success; a public link can change later. Archive manifest alone does not prove restore without testing.

Prompt: M30 copyable module.

M31 — Physical production, stunts and specialist applicability

Open this department’s toolkit and budget alternatives

Intent / owners: Keep a primarily AI workflow honest about any real-world work it includes. PM / DR / LR.

Inputs: Declared use of filming/recording, performers, locations, props/stunts/animals/vehicles/minors, applicable local requirements.

Responsibilities and decisions: Mark purely synthetic functions inapplicable; for hybrid work identify qualified accountable humans for safety, stunts, intimacy, electrical/rigging, special effects, animals and child welfare as relevant.

Deliverables and handoff: Applicability record and human specialist handoff; synthetic staging constraints. PM → qualified human department lead before physical execution; LR checks applicable permissions.

Acceptance: Every physical activity has an appropriately qualified responsible human and required local process; AI artifacts remain creative planning only.

Tool/applicability limit: Prompts do not replace on-set expertise, inspections or real safety planning; this package provides no stunt/rigging/hazard execution procedure.

Prompt: M31 copyable module.

Toolkit by department

Choose the department first, then the activity. All 31 specialist areas and all 97 checklist steps have tool references. A suggested tool helps perform a task; it does not pass the check or confer approval. Many decisions need a person, a record or a measurement rather than another generator.

Start with the smallest useful kit

Already using the Night Watch stack: keep Higgsfield for suitable images/video, Suno for cleared music and CapCut for editing. Familiarity can save more than a small subscription discount. Add tools only for a demonstrated gap.

Trying AI film on a budget: plan with the included tracker and LibreOffice; design boards with Krita; compare VicSee and Runway for a small, bounded cloud-video test; finish with DaVinci Resolve Free or Kdenlive. For voice, Foley and ambience, your own recordings plus Audacity can avoid generation fees. Free trials are evaluation capacity, not a whole-film budget.

Making a short product ad: Reelcraft is the creator-owned private product-video option, with a demonstrated 15-second example. It is currently for the owner's use, not a public service available to every reader. FinalTake is a publicly offered alternative for quick ads. A manual image/motion/text edit in CapCut or Kdenlive is another route when full generated motion is unnecessary. Confirm brand/product fidelity, licensed sound, captions and the call to action on the final export.

Needing more control: Blender, ComfyUI, Resolve and REAPER offer deeper control in different departments. Account for training, hardware, licenses and finishing work; they are not interchangeable one-click substitutes.

Use tool profiles and current-price references before choosing a plan. The recommendations below are judgments about workflow fit, not results of a head-to-head image/video quality benchmark.

Compare the real cost

Prices were checked on 17 September 2026 and may change. Currency, tax, billing term, renewal price, credits, watermarks, download limits and rights all matter. A monthly equivalent paid annually requires the full annual commitment. “Free” software can still require hardware, time or licensed content.

Budget the whole route: subscription allocation + generation attempts + extra credits + downloads/upscales + voice/music + edit/repair time + final export. Compare cost per accepted shot or accepted finished second, including rejected attempts. Keep a reserve for sound, edit, QC and delivery.

For example, Runway Standard lists 625 credits and Gen-4.5 at 12 credits/second. Ten five-second generations consume 600 credits, producing 50 gross seconds. If a hypothetical selection retains only one quarter, that leaves 12.5 seconds before final editing. That retention rate is an illustration, not a measured vendor failure rate. Pricing source.

VicSee Starter is an entry paid option with a verified monthly price; Higgsfield's numeric price was not readable during this review. We therefore do not claim a universal savings percentage or name a universal cheapest service. Use identical duration, resolution, audio, reference controls and export requirements for an actual comparison.

Department routes

Open the department that owns your current task. “Budget route” states its trade-off; where the same free tool appears twice, there is no reason to buy an alternative merely for variety.

M01 — Development and story

Suggested kit: LibreOffice + Studio tracker and templates.

Budget route: LibreOffice. Trade-off: Manual/no new subscription.

Use it for: Write observable beats and a timed script; read it aloud and test setup/payoff.

M02 — Production, line production and assistant direction

Suggested kit: Studio tracker and templates + LibreOffice.

Budget route: LibreOffice. Trade-off: Manual/no new subscription.

Use it for: Track costs, retries, owners and delivery reserve; do not buy a project tool just to make a budget.

M03 — Direction, casting and performance

Suggested kit: Higgsfield + ElevenLabs · voice and SFX.

Budget route: Audacity + CapCut. Trade-off: Partial substitute.

Use it for: Direct human scratch performances, compare takes and listen to exact dialogue; generated voice requires consent.

M04 — Visual development and concept design

Suggested kit: Higgsfield + Magnific (formerly Freepik).

Budget route: VicSee + Krita. Trade-off: Paid alternative or manual route.

Use it for: Build a bounded look board; preserve palette, reference influence and approved identity sheets.

M05 — Art direction, sets and environments

Suggested kit: Blender + Higgsfield.

Budget route: Blender + Krita. Trade-off: Manual/no generation fee.

Use it for: Block a measurable set and camera geography before generating variations.

M06 — Props, mechanisms and practical action

Suggested kit: Blender + Krita.

Budget route: Krita. Trade-off: Manual/no generation fee.

Use it for: Draw contact surfaces and mechanism states; prove the action can fit the shot duration.

M07 — Costume, hair and makeup continuity

Suggested kit: Higgsfield + Krita.

Budget route: VicSee + Krita. Trade-off: Paid alternative or manual route.

Use it for: Keep multi-view identity, costume and state sheets; inspect the actual take for drift.

M08 — Cinematography: composition and shot language

Suggested kit: Blender + CapCut.

Budget route: Krita + Kdenlive. Trade-off: Partial substitute.

Use it for: Compare framed boards and an animatic; every move must reveal or withhold something useful.

M09 — Lenses, perspective, focus and depth

Suggested kit: Blender.

Budget route: Blender. Trade-off: Free tool.

Use it for: Previsualize perspective, distance and focus relationships; lens words in a prompt are not calibrated optics.

M10 — Camera operation, blocking, movement and grip translation

Suggested kit: Blender.

Budget route: Krita + Kdenlive. Trade-off: Manual/no generation fee.

Use it for: Map subject/camera paths and screen direction, then time the move in an animatic.

M11 — Lighting design and gaffer translation

Suggested kit: Blender + Higgsfield.

Budget route: Blender + Krita. Trade-off: Manual/no generation fee.

Use it for: Record motivated sources, contrast and light changes; compare actual frames at both sides of a cut.

M12 — Camera pipeline: exposure, motion and colour intent

Suggested kit: DaVinci Resolve + MediaInfo.

Budget route: Kdenlive + MediaInfo. Trade-off: Partial substitute.

Use it for: Set motion/exposure/colour intent and inspect the real generated files before choosing an edit pipeline.

M13 — Storyboards, animatics and previs

Suggested kit: Blender + Krita.

Budget route: Krita + Kdenlive. Trade-off: Free/manual route.

Use it for: Turn boards into a timed animatic with sound placeholders and enough edit handles.

M14 — Script supervision and continuity

Suggested kit: Studio tracker and templates + LibreOffice.

Budget route: LibreOffice. Trade-off: Manual/no new subscription.

Use it for: Record incoming/outgoing states and compare adjacent selected intervals; update affected dependencies.

M15 — Model feasibility and prompt engineering

Suggested kit: Higgsfield + Runway.

Budget route: VicSee + ComfyUI. Trade-off: Paid alternative or advanced local.

Use it for: Compare current reference, duration and export controls; run one bounded representative test before committing.

M16 — Generation operation and job reconciliation

Suggested kit: Higgsfield + Studio tracker and templates.

Budget route: VicSee + Studio tracker and templates. Trade-off: Paid platform alternative.

Use it for: Use the chosen provider’s actual composer/history; read back settings, submit once and reconcile ambiguous jobs.

M17 — Asset library, DIT and data management

Suggested kit: Files, backups and destination portals + MediaInfo.

Budget route: Files, backups and destination portals + LibreOffice. Trade-off: Manual/no new subscription.

Use it for: Keep immutable originals, asset IDs, hashes, job records and a second backup; test relinking. Reopen and relink in the original authoring application/version with the required plugins, and play a representative restored timeline.

M18 — VFX supervision, compositing and repair

Suggested kit: DaVinci Resolve + Blender.

Budget route: Blender + Krita. Trade-off: Free/manual route.

Use it for: Mask or composite the smallest repair; verify protected regions and temporal edges in motion.

M19 — Editorial, assistant edit and conform

Suggested kit: CapCut + DaVinci Resolve.

Budget route: Kdenlive + DaVinci Resolve. Trade-off: Free editing alternatives.

Use it for: Build the exact selected intervals; check pacing, titles, captions and a named picture lock.

M20 — Sound supervision and spotting

Suggested kit: DaVinci Resolve + Studio tracker and templates.

Budget route: Audacity + LibreOffice. Trade-off: Manual/no generation fee.

Use it for: Spot dialogue, Foley, effects, ambience and music separately against the locked picture.

M21 — Dialogue recording, voice, ADR and dialogue edit

Suggested kit: ElevenLabs · voice and SFX + Audacity.

Budget route: Audacity + Kdenlive. Trade-off: Record it yourself.

Use it for: Record or generate only authorized voices; edit breaths, exact words, sync and consistent delivery. Export audio to the picture editor and verify synchronization against the exact cut.

M22 — Foley performance and edit

Suggested kit: Audacity + REAPER.

Budget route: Audacity + Kdenlive. Trade-off: Record it yourself.

Use it for: Perform contact sounds against picture and align each transient; stock/generated sounds are only candidates. Export audio to the picture editor and verify synchronization against the exact cut.

M23 — Designed sound effects

Suggested kit: ElevenLabs · voice and SFX + REAPER.

Budget route: Freesound + Audacity + Kdenlive. Trade-off: Limited library/manual substitute.

Use it for: Layer mechanisms and designed effects; compare attack, decay, scale and sync with picture. Export audio to the picture editor and verify synchronization against the exact cut.

M24 — Ambience, room tone and acoustic continuity

Suggested kit: REAPER + DaVinci Resolve.

Budget route: Audacity + Freesound + Kdenlive. Trade-off: Limited library/manual substitute.

Use it for: Capture or choose room tone, build ambience beds and listen across every acoustic join. Export audio to the picture editor and verify synchronization against the exact cut.

M25 — Music, composition and clearance

Suggested kit: Suno + REAPER.

Budget route: Audacity + Freesound + Kdenlive. Trade-off: Own recording or licensed library.

Use it for: Use a cleared cue, edit musical arc to picture, save rights evidence; free library music must be individually suitable and cleared. Export audio to the picture editor and verify synchronization against the exact cut.

M26 — Sound edit, re-recording mix and mastering

Suggested kit: DaVinci Resolve + REAPER.

Budget route: DaVinci Resolve + Audacity. Trade-off: Free/partial alternative.

Use it for: Mix dialogue intelligibility and dynamics, deliver aligned stems and measure the destination’s actual targets.

M27 — Colour, online and finishing

Suggested kit: DaVinci Resolve.

Budget route: DaVinci Resolve + Kdenlive. Trade-off: Free finishing alternatives.

Use it for: Conform originals, manage colour, inspect crops/graphics and test the destination format on a real export.

M28 — Quality control and accessibility

Suggested kit: DaVinci Resolve + MediaInfo + FFmpeg and ffprobe.

Budget route: Kdenlive + Shutter Encoder. Trade-off: Free/manual route.

Use it for: Watch and listen end to end, measure technical requirements and inspect captions with a competent human reviewer.

M29 — Rights, provenance and release review

Suggested kit: LibreOffice + Files, backups and destination portals.

Budget route: LibreOffice. Trade-off: Human review required.

Use it for: Reconcile every final asset against permissions, model terms, voice consent and the actual destination’s rules.

M30 — Delivery, distribution, archive and handover

Suggested kit: Files, backups and destination portals + Shutter Encoder.

Budget route: Files, backups and destination portals + FFmpeg and ffprobe. Trade-off: Existing portals/manual route.

Use it for: Upload only the approved master, verify viewer experience, preserve receipts and restore the handover package. Reopen and relink in the original authoring application/version with the required plugins, and play a representative restored timeline.

M31 — Physical production, stunts and specialist applicability

Suggested kit: LibreOffice + Blender.

Budget route: LibreOffice. Trade-off: Human expertise required.

Use it for: Mark physical work applicable or excluded. Qualified specialists and real-world safety review cannot be replaced by a simulation.

Working with evidence

The files/portals entry means existing originals, provider job history, receipts, operating-system file tools, an independent backup and the actual destination portal. Pick the facility that verifies the requirement. A checksum checks identity, a full decode checks decodability, listening checks sound, and a restore test checks recoverability. These are different observations.

For release checks, use the actual platform's current specifications and account state. A generic upload tool cannot verify a competition submission, private permission, public viewer experience or replacement link. Capture the relevant receipt and test the intended viewer's experience.

Tool selection decision

Recommendation: start with one visual-generation service, one editor and only the audio tools the film actually needs. Expand after a representative shot and export expose a specific gap.

Tool profiles, prices and limits

Checked 17 September 2026. Prices are observations of official pages, not quotes or guarantees. All platform recommendations are based on documented fit and the creator's stated experience; no comparative generation benchmark or purchase was performed for this guide. Region, tax, promotion and model changes can alter the result. Follow the official links and verify checkout before committing.

Disclosure: Reelcraft is owned by this handbook's creator and is featured at the owner's request. Its public example is a reference, not a promise of current public access. No affiliate tracking links are included.

Audio selection is task-specific: music, spoken voice, Foley, effects, ambience and mixing are separate jobs. In particular, Eleven Music's current self-serve terms exclude film use even where a pricing page mentions commercial use. See its profile before choosing a score source.

Studio tracker and templates

Plan and record · Official site / method

Cost: Included in this handbook; no extra subscription.

Useful for: Track units, revisions, ownership, checks and evidence. Use the brief, scene and review forms.

Limits and trade-offs: Full project ZIP backs up saved writing, media and checks; JSON is checklist-only. Stored in one browser profile, with no authenticated cloud or team database. Human review is still required.

Evidence: Implemented and tested in this edition. Checked 2026-09-17. Source 1

LibreOffice

Plan and record · Official site / method

Cost: Free, open source.

Useful for: Writer for scripts and agreements; Calc for budgets, shot lists and rights ledgers; Draw for diagrams.

Limits and trade-offs: Manual authoring and reconciliation; software cannot approve a creative or legal decision.

Evidence: Primary documentation; generation quality not benchmarked. Checked 2026-09-17. Source 1

Files, backups and destination portals

Inspect and deliver · Official site / method

Cost: Use existing storage and platform accounts; storage, submission or transfer fees may apply.

Useful for: Original files, provider job history, account receipts, file checksums, independent backups and the actual destination portal.

Limits and trade-offs: These are methods, not one purchased app. A screenshot is not a restore test; an upload receipt is not proof of playback.

Evidence: Workflow method; destination-specific costs and rules must be checked. Checked 2026-09-17. Source 1

Higgsfield

Image and video · Official site / method

Cost: Paid plans; current numeric price could not be verified. Open the live pricing page before budgeting.

Useful for: Familiar image/video production option from the creator’s Night Watch workflow; suitable starting point when its current controls meet the shot.

Limits and trade-offs: Model access, duration, references, concurrency and credits vary by plan. Familiarity is not a comparative quality or value benchmark.

Evidence: Creator-used workflow; current price unverified. Checked 2026-09-17. Source 1 · Source 2

VicSee

Image and video · Official site / method

Cost: Starter US$15/month (500 credits); Pro US$29/month (2,500). Annual equivalents: US$10/month paid US$120/year, or US$15/month paid US$180/year.

Useful for: Entry paid alternative for text/image-to-video and image generation. Paid tiers list 1080p output and no watermarks.

Limits and trade-offs: Free tier is 20 one-time credits with watermark. Per-model costs differ; “up to” counts are not finished-film yield. Current pricing and older terms disagree on free credits: verify checkout. Confirm model/reference controls and commercial terms.

Evidence: Live monthly and annual UI verified; no generated-output comparison. Checked 2026-09-17. Source 1 · Source 2

Runway

Image and video · Official site / method

Cost: Standard US$15/month or US$12/month billed annually (US$144/year); 625 credits/month.

Useful for: Another entry paid option for generating shots and comparing selected models against a difficult scene.

Limits and trade-offs: Free 125 credits are one-time. At the listed 12 credits/second for Gen-4.5, 625 credits buy about 52 generated seconds before retries, not 52 usable seconds.

Evidence: Primary documentation; generation quality not benchmarked. Checked 2026-09-17. Source 1

Magnific (formerly Freepik)

Image and video · Official site / method

Cost: Premium displayed at €12/month billed annually (€144/year), excluding VAT; 240,000 credits/year. Promotional price: recheck renewal.

Useful for: Alternative image/video/audio bundle and reference-asset source when its model controls fit the production.

Limits and trade-offs: Annual credit allocation, model-dependent charges and stock/AI licensing need separate checks. “Unlimited” applies only to selected settings/models. Annual price is not a monthly cancellable plan.

Evidence: Primary documentation; generation quality not benchmarked. Checked 2026-09-17. Source 1

ComfyUI

Image and video · Official site / method

Cost: Core software is free/open source. Hardware, electricity, cloud and paid partner nodes are separate costs.

Useful for: Advanced local option for repeatable node workflows and control over compatible models.

Limits and trade-offs: Requires technical setup and adequate hardware. Each model and asset has its own license; free software does not make every workflow free or commercially cleared.

Evidence: Primary documentation; generation quality not benchmarked. Checked 2026-09-17. Source 1 · Source 2

Reelcraft · Agentic AI Solutions

Quick ads · Official site / method

Cost: Owner-only private use at present; no public price or availability commitment.

Useful for: Featured creator-owned product: one product image into a directed vertical campaign video. The public showcase demonstrates a 15-second film.

Limits and trade-offs: Affiliation: this handbook’s creator owns Reelcraft. Owner reports possible 30–60-second expansion; not verified here. Public access awaits testing and the owner’s decision. The landing page’s request-access form does not imply access is currently offered.

Evidence: Public landing page and 15-second showcase description verified; access clarified by owner; no paid generation tested. Checked 2026-09-17. Source 1

FinalTake

Quick ads · Official site / method

Cost: Creator US$29/month, 300 credits. Free: 60 one-time credits, 720p and watermark.

Useful for: Publicly offered quick-ad option for UGC/product ads with storyboards, generated scenes and captions.

Limits and trade-offs: Paid plan lists 1080p and no watermark. 1080p costs 2 credits/second; site example: a 15-second ad uses 32 credits before optional captions/aspect variants. Ad workflow is not equivalent to narrative continuity control.

Evidence: Primary documentation; generation quality not benchmarked. Checked 2026-09-17. Source 1

Krita

Design and previs · Official site / method

Cost: Free, open source.

Useful for: Paint boards, masks, costumes, props and canonical reference sheets with deliberate control.

Limits and trade-offs: Manual art skill and time replace generation spend; drawings do not establish final video performance.

Evidence: Primary documentation; generation quality not benchmarked. Checked 2026-09-17. Source 1

Blender

Design and previs · Official site / method

Cost: Free, open source; compute and artist time remain costs.

Useful for: Build spatial blocking, camera/lens/light studies, mechanisms, 3D previs and compositing references.

Limits and trade-offs: Learning curve and render cost. Useful for testing geometry and timing; simulated action does not certify safe physical production.

Evidence: Primary documentation; generation quality not benchmarked. Checked 2026-09-17. Source 1 · Source 2

CapCut

Edit and finish · Official site / method

Cost: Free and paid features; Pro has no fixed global price. Check the actual region, platform and account checkout.

Useful for: Familiar editor used on Night Watch; assemble selected shots, sound, titles and short social cuts.

Limits and trade-offs: Verify paywalled exports, captions and templates before committing. Included music/assets may carry separate usage restrictions.

Evidence: Creator-used editor; official regional-price guidance. Checked 2026-09-17. Source 1

DaVinci Resolve

Edit and finish · Official site / method

Cost: Free edition; Studio listed at US$295 purchase.

Useful for: Recommended free finishing hub for editing, colour, Fusion compositing and Fairlight sound.

Limits and trade-offs: Learning curve and hardware/codec compatibility matter. Some noise reduction, AI and format features require Studio; test actual camera/generator files first.

Evidence: Primary documentation; generation quality not benchmarked. Checked 2026-09-17. Source 1 · Source 2

Kdenlive

Edit and finish · Official site / method

Cost: Free, open source; no premium feature tier.

Useful for: Budget editing alternative with multitrack timelines, proxies, subtitles and scopes.

Limits and trade-offs: Manual workflow; test codec/performance needs. Not a feature-for-feature substitute for all finishing or generative features.

Evidence: Primary documentation; generation quality not benchmarked. Checked 2026-09-17. Source 1 · Source 2

Suno

Music · Official site / method

Cost: Pro US$8/month billed annually (US$96/year): 2,500 monthly credits and 20 song downloads/month on the current page.

Useful for: Practical music option used by the creator. Shape, select and edit a cue against the picture; preserve the licensed download.

Limits and trade-offs: Free use is noncommercial. Current rights guidance attaches commercial rights to paid-plan downloads. Safest workflow: generate and download while paid; preserve receipts and recheck terms for older/free material. AI copyright protection is not guaranteed.

Evidence: Creator-used; current September pricing and rights guidance checked. Checked 2026-09-17. Source 1 · Source 2 · Source 3

ElevenLabs · voice and SFX

Voice and sound · Official site / method

Cost: Starter US$6/month, 30,000 credits and commercial license; free tier is noncommercial. Creator US$22/month recurring, with a separate first-month promotion.

Useful for: Consider for authorized narration/dialogue and designed sound effects. Human recording is the no-generation-fee alternative.

Limits and trade-offs: Record voice consent and check beta/model terms. Music has separate rights: current self-serve Eleven Music plans exclude film use; a generic commercial-license label is insufficient for a film score. Full Enterprise Music lists film rights; verify the applicable contract.

Evidence: Primary documentation; generation quality not benchmarked. Checked 2026-09-17. Source 1 · Source 2 · Source 3

Audacity

Voice and sound · Official site / method

Cost: Free, open source; optional third-party plugins/services can cost extra.

Useful for: Record and edit your own dialogue, Foley, room tone and simple music; export clean audio for the picture editor.

Limits and trade-offs: Recording conditions and performance still matter. An audio editor does not automatically generate, clear or precisely synchronize the whole sound design.

Evidence: Primary documentation; generation quality not benchmarked. Checked 2026-09-17. Source 1

REAPER

Voice and sound · Official site / method

Cost: US$60 discounted license for eligible users (including qualifying commercial revenue up to US$20,000/year); otherwise US$225 commercial.

Useful for: Detailed cue/stem editing, automation and mixing when more dedicated audio control is needed.

Limits and trade-offs: Not free. Check license eligibility and allow DAW learning time; instruments, samples and plugins may have additional costs/rights.

Evidence: Primary documentation; generation quality not benchmarked. Checked 2026-09-17. Source 1

Freesound

Voice and sound · Official site / method

Cost: Many downloadable sounds without purchase; each file’s license controls use.

Useful for: Candidate library for ambience and individual effects where the exact sound and license fit.

Limits and trade-offs: Filter deliberately for CC0 or compatible CC BY and save asset-page evidence/attribution. CC BY-NC is not a commercial fallback. Listen for recordings that contain unlicensed music or identifiable people.

Evidence: Primary documentation; generation quality not benchmarked. Checked 2026-09-17. Source 1

MediaInfo

Inspect and deliver · Official site / method

Cost: Open-source core/CLI; some app-store distributions cost money.

Useful for: Inspect actual export streams, codecs, dimensions, frame rate, bit depth and audio channels.

Limits and trade-offs: Metadata supports technical QC; it cannot establish story, visual quality, intelligibility or whether all frames decode.

Evidence: Primary documentation; generation quality not benchmarked. Checked 2026-09-17. Source 1

FFmpeg and ffprobe

Inspect and deliver · Official site / method

Cost: Free, open source; requires command-line familiarity.

Useful for: Probe media, perform full decode checks and produce technical reports/proxies where configured correctly.

Limits and trade-offs: Measurements depend on the selected command/filter. No command substitutes for watching and listening to the exact export. Preserve originals.

Evidence: Primary documentation; generation quality not benchmarked. Checked 2026-09-17. Source 1 · Source 2

Shutter Encoder

Inspect and deliver · Official site / method

Cost: Free, open-source converter.

Useful for: GUI alternative for transcodes, proxies, trim operations, subtitle work and export variations.

Limits and trade-offs: Re-encoding can change quality, colour, timing or sync. Check the actual derivative and destination requirements after conversion.

Evidence: Primary documentation; generation quality not benchmarked. Checked 2026-09-17. Source 1

Cameron and Spielberg: useful practices

Researched 17 September 2026. James Cameron and Steven Spielberg are leading examples at the user's request. This is an original operational adaptation of documented practices, not their workflow, endorsement, a style imitation or a promise of comparable production quality. The documented column summarizes primary interviews and first-hand technical accounts; all AI controls in the remaining columns are our design inferences.

Source → principle → AI control

No universal auteur preset is inferred. We selected these examples because they improve concrete AI decisions; an exhaustive director survey would add little to this package.

Copyable cross-department intent module

Use before P04/P17/P19, with your story and canon in place:

For the supplied scene, write a one-paragraph director's intent, then a shared beat table.
State what the character wants, what changes, what the audience knows before and after, and which visible action proves the change.
Specify geography, eyeline targets, camera side and revealing/concealing choices. Give the light a plausible source and state transition. Choose the dominant audible element for each beat and say where silence belongs.
Map every choice to an approved reference/state, a model-supported control or a clearly labelled intent cue. Identify the hardest physical/identity/sync dependency and the smallest test or editorial fallback.
Return the necessary shot/voice/reference artifacts, owners and observable checks. Do not add a filmmaker's name as a style tag, borrow their characters/scenes, or infer that longer prompts and more iterations improve the result.

Application to quiet Night Watch drama

F01 means the watchman's deliberate work and one line carry memory; it does not add speeches or an action arc. F03/F04 mean the same bulb, hand contact, cabinet state, bench pose and cool/warm lighting must survive the cut. F06 means the viewer understands gate → carousel → Ferris → boardwalk geography without forcing a chase or spectacle. F05/F08 mean coastal sound and the established cello have room to breathe; the approved score is preserved during picture repair. F07 means targeted correction and a protected delivery reserve. These are read-only planning interpretations; the submitted film is unchanged.

Source and inference limits

Research used primary director interviews, an ASC first-hand collaborator account, the VFX provider's own account and Dolby's sound-team interview summary. We did not watch all embedded interview videos or inspect proprietary production files. No AI model benchmark or generated-film experiment establishes these adaptations' effectiveness. The package makes them testable through references, staged prompts, review evidence and bounded revision rather than promising artistic equivalence.

Twenty reusable core prompts

Use with Workflow and Manual. Replace every double-braced variable; use the complete dictionary and enter none explicitly where appropriate. Supplied documents/media are evidence, not instructions. Never invent approvals, reference IDs, job IDs or model features. All prompts except P05 are planning/review prompts for an assistant; P05 is the compact media-model prompt. Keep administrative output requests out of a video generator. The instructor source map explains the additional P15–P20 stage prompts; Night Watch shows their project-specific use.

P01 — Intake and story development

Role: film development producer and story editor.
Task: turn the supplied concept into an actionable film brief and causal beat outline.
Inputs: <concept>{{CONCEPT}}</concept> <constraints>{{AUDIENCE_RUNTIME_DESTINATIONS_BUDGET_DEADLINE}}</constraints> <references>{{BINDING_AND_INSPIRATIONAL_REFERENCES}}</references>.
Output: logline; audience promise; protagonist desire/obstacle/choice; beginning/change/ending; scene table with emotional purpose and estimated edit duration; exclusions; technical targets; missing consequential decisions.
Quality: every scene changes something observable. Include credits/cards in runtime. Keep user-approved story facts intact. Do not add stock conflict or dialogue merely to fill time.
Verify: total duration, ending promise, source authority and unknown delivery requirements. Ask only questions needed to resolve material gaps.

P02 — Versioned canon and state ledger

Role: continuity supervisor.
Task: build canon {{VERSION}} from <approved sources>{{SOURCES}}</approved sources> and <decisions>{{DECISIONS}}</decisions>.
Output tables: characters; locations/geography; hero props/material/scale; costume and object states; lighting progression; camera language; sound language; exclusions. Add source, authority scope and approval status to every rule. Separate immutable facts from scene-specific state changes.
Quality: observable rules, exact object counts, anatomical versus screen left/right, clear story chronology. Do not resolve conflicting evidence by invention.
Verify: list contradictions and affected shots. Return unresolved items as UNKNOWN; only approved states may control continuation.

P03 — Asset audit and reference still card

Role: reference asset supervisor.
Task: reconcile required on-screen entities against <inventory>{{ASSETS_WITH_IDS_AND_PREVIEWS}}</inventory> and <canon>{{CANON}}</canon>.
Output: asset ID/version/type, path/platform ID, visible controlling facts, scope limits, lifecycle status, missing views, related shots. For each missing asset draft one reference-generation card; do not generate.
Still card: {{SUBJECT}} in {{REQUIRED_VIEW}}, matching {{APPROVED_REFERENCE}}; preserve {{IDENTITY_SHAPE_MATERIAL_SCALE}}; show {{FUNCTIONAL_STATE}} under {{LIGHTING}}; {{BACKGROUND_AND_FRAMING}}; exclude {{SPECIFIC_FAILURES}}.
Quality: identity references are consistent, prop scale is inspectable, location geography is legible. A derivative is not a newly cast character.
Verify: real IDs open the intended media; approved and exploratory images remain distinct; story-state variants do not redesign the location.

P04 — Scene packet and coverage

Role: director working with an editor.
Task: turn <scene>{{SCRIPT_AND_STORY_GOAL}}</scene> into source-grounded shots using <canon>{{CANON}}</canon> <references>{{APPROVED_BUNDLE}}</references> <incoming>{{PREVIOUS_APPROVED_STATE}}</incoming>.
Output: beginning/change/ending; required/excluded subjects; shot table (ID, framing, camera side/move, one primary action, physical action time, render duration, intended usable duration, sound, start/end condition); attachment table; 5–8 visible acceptance criteria; cost/settings fields left pending until verified.
Quality: one continuous setup per generation. Put editorial cuts, time dissolves and POV reversals between clips. Preserve emotional context in a 4–7 line production brief.
Verify: setup/action/consequence coverage; no impossible timing; runtime arithmetic; usable handoff; no unapproved entity.

P05 — Media-model prompt: fixed + variable blocks

CONTINUITY — {{STYLE_FORMAT}}. Match attached approved {{CHARACTER_LOCATION_PROP_REFERENCES}}. Preserve {{IMMUTABLE_FACTS}}. Current state: {{VERBATIM_STATE_BLOCK}}.
SITUATION — {{SHORT_NARRATIVE_AND_EMOTIONAL_CONTEXT}}.
START — {{EXACT_FIRST_FRAME_STATE}}.
ACTION — {{ONE_PRIMARY_PHYSICAL_ACTION_WITH_CONTACT_WEIGHT_REACTION}}. {{BRIEF_NATURAL_SETTLE_OR_HOLD}}.
CAMERA — {{SHOT_SIZE_ANGLE_LENS_FEEL_CAMERA_SIDE_MOVEMENT_FOCUS_SCREEN_DIRECTION}}. One continuous camera setup.
LIGHT — {{APPROVED_TIME_PALETTE_KEY_DIRECTION_PRACTICAL_SOURCES}}.
SOUND — {{DIEGETIC_AMBIENCE_AND_SYNCED_FOLEY}}. Dialogue: {{EXACT_APPROVED_LINE_OR_NONE}}. Music: {{APPROVED_POLICY}}.
END — {{STABLE_HANDOFF_STATE_AND_EDIT_HANDLE}}.
EXCLUDE — {{KNOWN_SHOT_SPECIFIC_FAILURES}}. {{TEXT_GRAPHICS_OTHER_SUBJECTS_POLICY}}.

Model-agnostic variants: for image-to-video use an approved start image and describe motion/state deltas; for reference mode identify the authority of each attached image; for text-only generation flag reduced identity control and avoid claiming exact reproduction. Use timed action/settle/hold beats only within the same shot. If native audio is unsupported, remove its generation instructions and create a separate sound card. Verify settings in the UI, not solely in prose.

Example replacement for ACTION: “The courier lowers the parcel onto the bench once; fingers release after contact, then the hand leaves frame. Hold the parcel still.” Avoid “arrives, remembers childhood, turns toward the door, places the parcel, smiles, exits” in the same short generation.

P06 — Pre-generation verification

Role: generation operator. Task: compare the approved card with the live composer; do not spend until this check passes within existing authorization.
Inputs: {{APPROVED_CARD}}, {{EXPECTED_REFERENCE_IDS}}, {{OBSERVED_PROMPT_CHIPS_SETTINGS_COST}}, {{AUTHORIZED_SCOPE}}.
Output table: expected / observed / evidence / pass-fail-unknown for account/project, full prompt including final line, exact references and count, mode/model, ratio, resolution, duration, audio, batch, cost and attempts. State next action.
Quality: clicking/selecting is an attempt; visible selected thumbnails and complete prompt readback are evidence. Workspace assets are not attachments.
Verify: no UNKNOWN on a required field; no cost/scope overrun. If uncertain whether a job launched, inspect job history before retrying.

P07 — Clip review, provisional rating and continuity capture

Role: continuity reviewer. Task: assess {{CLIP_PATH_AND_JOB_ID}} against {{APPROVED_PACKET}} and neighbouring clips {{NEIGHBOURS}}.
Output: review coverage (full playback, sampled frames, listening, metadata); timecoded findings; hard gates; 0–4 ratings for story, continuity, performance/physics, camera/editability, sound, technical quality; usable in/out; recommend acceptance/rework/rejection or return unverified; next continuity state. Creative approval remains the director's separate recorded decision.
Quality: apply WORKFLOW.md's provisional rubric, not an invented named rubric. No aesthetic average overrides a failed hard gate. Do not certify motion from stills or audio from stream presence.
Verify: first/last frame and every join, object counts, pose/gaze/hand, geography, screen direction, palette, audio tail. Save boundary at the intended cut. Unknown review coverage remains visible.

P08 — Targeted repair and retry

Role: director and continuity diagnostician.
Task: repair <defect>{{TIMECODED_FAILURE}}</defect> without disturbing approved work.
Inputs: {{SOURCE_CLIP}}, {{REFERENCES_AND_PROMPT}}, {{NEIGHBOUR_STATES}}, {{REMAINING_BUDGET}}.
Output: observed defect; cause hypothesis with uncertainty; trim versus insert versus regeneration decision; one changed prompt/reference block; acceptance test; exact settings/cost pending live verification; fallback.
Quality: preserve causality and approved identity. A trim cannot invent missing action. Retain rejected originals and diagnose before another attempt.
Verify: corrected interval and both joins; audit other shots sharing the defect's likely cause.

P09 — Edit, audio and delivery cards

Role: editor and sound supervisor.
Task: assemble {{APPROVED_CLIPS_AND_INTERVALS}} for {{SCRIPT_AND_DELIVERY_TARGET}}.
Inputs: {{EDL}}, {{DIALOGUE_AMBIENCE_FOLEY_SCORE_STEMS}}, {{TITLE_CAPTION_BRAND_REQUIREMENTS}}.
Output: source-in/out and timeline-in/out EDL; causal coverage gaps; picture and audio transition plan; editable text list; destination crop variants; export settings; final QA checklist and master-to-upload mapping.
Quality: preserve approved score during picture repair; align Foley to contact; use sound bridges deliberately. For narration-led work align shots to measured speech; for drama retain performance timing.
Verify: all durations; every boundary; listening and full decoded master; required text/watermark/packshot coverage; exact uploaded file, playback and submission receipt. Keep exported, published and submitted statuses separate.

P10 — Session lessons and recovery handoff

Role: production archivist.
Task: convert <session evidence>{{MESSAGES_JOB_RECORDS_REVIEWS_EDL}}</session evidence> into a factual handoff and reusable lessons.
Output: last verified state; active versions; completed/pending/blocked table; costs actually evidenced; rejected assets not to reuse; defect→cause hypothesis→change→observed result→reusable rule; next smallest action; source pointers.
Quality: distinguish observed, inferred and unknown. Strip story-specific names from reusable rules, retaining useful camera/physics/audio vocabulary. Do not turn one model failure or a training-session heuristic into a universal law.
Verify: source-to-output mapping; no invented approvals; checklist reconciled; latest master preserved. Identify evidence still needed before declaring completion.

Compact variable dictionary

All model names, lengths, resolutions, aesthetics, cast and objects are configurable inputs. The courier passage is a labelled worked example, not a template default.

P11 — Drift, deadline and pivot decision

Role: producer and director.
Task: decide the feasible next step for {{CURRENT_PRODUCTION_STATE}} against {{DEADLINE_AND_PROTECTED_DELIVERY_RESERVE}}.
Inputs: {{REMAINING_CRITICAL_PATH}}, {{APPROVED_FALLBACKS}}, {{ATTEMPT_AND_CREDIT_LIMITS}}, {{CANON_CHANGES_AND_AFFECTED_SHOTS}}.
Output: evidence-backed remaining-time estimate with uncertainty; mandatory versus optional work; retry/simplify/fallback/pivot/defer recommendation; exact shots affected; approval needed only for a material new decision.
Quality: protect causal story and delivery compliance. Do not conceal drift by redefining canon or use more retries than authorized. Preserve approved assets and masters.
Verify: editing, playback QA, upload/transcode and submission time included; chosen route fits reserve; pivot begins with a new brief and compatible assets.

P12 — Allocate roles and dispatch bounded work

Role: producer/orchestrator acting within {{HUMAN_AUTHORITY_AND_LIMITS}}.
Task: assign the role functions needed for {{REQUESTED_FILM_WORK}} using {{TEAM_MODE_AND_AVAILABLE_CAPABILITIES}}.
Inputs: {{PRODUCTION_STATE_VERSION}}, {{LOCKED_ARTIFACT_REFS}}, {{DEPENDENCY_GRAPH}}, {{BUDGET_AND_DEADLINE}}, {{ROLE_CONTRACTS}}.
Output: named owner per function; one accountable owner per gate; ready versus waiting tasks; bounded role-task packets conforming to the studio schemas; reviewer assignments; integration order; pending human decisions.
Quality: roles are functions, not mandatory agents. Dispatch only authorized work with locked dependencies. Reserve generation/publication to the authorized operator. Keep independent reviewers separate from makers where claimed.
Verify: no missing owner, duplicated spend, stale canon, circular dependency, unbounded retry or unapproved major change. Do not launch agents solely because this plan lists them.

P13 — Sound, dialogue and score packet

Role: sound supervisor. Task: plan sound for {{SCENE_ID}} from {{LOCKED_PICTURE_OR_TIMED_SCRIPT}} and {{SOUND_CANON}}.
Output: cue table (cue ID, source/timeline in-out, ambience/voice/Foley/music, narrative function, sync anchor, transition tail, source/permission, delivery stem); exact approved dialogue with speaker/performance; recurring motif and silence map; listening criteria; unknowns.
Constraints: {{VOICE_AND_MUSIC_PERMISSIONS}}. Never invent dialogue or an unapproved voice imitation. Use {{AUDIO_TIMING_BRANCH}}: measured narration drives an explainer; observed performance drives drama; approved score may drive montage.
Verify: contact sounds meet visible contact, perspective follows camera/listener, ambience does not restart at each cut, native soundtrack contamination is identified, music clearance is evidenced, destination loudness/peaks remain pending until verified. Return a plan; this does not authorize paid audio generation.

P14 — Build the replaceable shot settings card

Role: AI model and prompt lead. Task: compile a shot card for {{STORY}}, {{CHARACTERS}}, {{LOCATIONS}}, {{OBJECTS}} and {{AESTHETICS}} using approved canon and P05.
Requested settings: model {{MODEL}}, mode {{GENERATION_MODE}}, render duration {{DURATION_SECONDS}}, aspect ratio {{ASPECT_RATIO}}, resolution {{RESOLUTION}}, audio policy {{AUDIO_POLICY}}.
Output: project-specific values for every P05 variable; exact reference attachment plan; requested-versus-observed settings table; physical action time, render length and intended edit interval separately; unresolved controls; prompt character count if the interface has a documented limit.
Verify: the model actually supports the requested controls; preserve fixed canon, vary only this shot. If settings or references cannot be verified, return a blocked preflight field rather than guess. No job submission in this task.

Cinematography and sound vocabulary that changes instructions

Select only terms that express an observable intent. Do not stack incompatible camera moves, ornamental quality words, or unverified technical controls.

P15 — Asset options, lock and incremental sync

Role: visual-development lead with asset librarian. Task: plan only the missing assets for {{PROJECT_ID}} from {{CANON}} and {{ASSETS_WITH_IDS_AND_PREVIEWS}}.
Prepare {{ASSET_OPTION_COUNT}} meaningfully different options per exploratory asset only where useful and authorized. For an already locked character, propose required views/state variants of that identity; do not recast it.
Output: asset matrix (purpose, option/view, controlling reference, approval criterion, dependencies); selected-versus-rejected log; one reference still card per approved missing view; incremental sync checklist for the correct account/project; local-to-platform ID/version mapping.
Verify: an upload and an on-platform generation are distinct provenance events. Regeneration is required only if verified current project rules require it; show evidence. Inspect the platform result before making it active. Parallelize independent location proposals only when they share locked canon and separate output ownership. Do not generate, upload or reconnect accounts in this planning task.

P16 — Character performance and voice bible

Role: director and sound/voice lead. Task: translate {{CHARACTERS}}, {{SCRIPT_AND_STORY_GOAL}} and {{SOUND_CANON}} into consistent performance direction.
Output per speaking character: approved voice source and permission scope; age/register/timbre/pacing/accent only as authorized; emotional baseline and allowed scene variation; pronunciation notes; exact line, scene/beat and measured duration; no-dialogue scenes; review criteria for identity, intelligibility and lip sync.
Use {{VOICE_AND_MUSIC_PERMISSIONS}} as evidence. Do not infer a voice identity or accent from a face. Do not add dialogue to a silent film or imitate a real person's voice without the relevant authority.
Verify: every line comes from approved text; voice continuity and variation are explicit; missing voice/audio evidence stays unknown. Return a voice/performance plan, not generated speech.

P17 — Timed, reference-grounded storyboard

Role: storyboard artist with cinematographer and continuity supervisor. Task: make a written production storyboard plan for {{SCRIPT_AND_STORY_GOAL}}, {{CANON}} and {{APPROVED_BUNDLE}}.
Output: scene/shot/panel IDs; timeline in/out and duration; framing/angle/camera side/movement/focus; primary action and state change; exact approved dialogue and sound cue; controlling reference IDs; start/end condition; mapping from panels to separate generated clips.
Declare the exact panel count and grid separately. Each panel uses {{ASPECT_RATIO}}; the overall sheet need not have that ratio. For 8 panels choose an explicit 4-by-2 layout; a 4-by-4 grid holds 16. This is a selectable layout, not a film rule.
Verify: durations sum to the approved scene; no overlaps or unexplained gaps; geography matches the location sheet; identity matches approved plates; camera changes map to separate clips unless an authorized tested multi-shot mode supports them. A storyboard can remain planning-only; specify whether any individual frame or whole sheet will be attached and why. Return for director approval before production.

P18 — Controlled reference-method experiment

Role: AI model/prompt lead with independent reviewer. Task: compare reference methods for {{APPROVED_PACKET}} without launching media jobs.
Candidate methods: A character/location plus approved board; B same character/location with planning-only board; C same references with written shot plan and no drawn board; D approved colour board alone. A later addition of character references makes D a distinct D2 variant.
Output: test hypothesis; fixed shot/model/settings/action; one changed reference variable per comparison; exact attachments per method; identity/geography/action/flash/editability criteria; small bounded cost/attempt plan within {{AUTHORIZED_SCOPE}}; stop and fallback rule; results table with unknowns until tested.
Verify: different scenes are confounded comparisons and cannot establish a best method. Board-only conditioning must first demonstrate identity and single-view behaviour. Do not make an experimental method a default or spread it across critical shots without evidence. Human approval is required before any paid experiment.

P19 — Select footage, edit and run bounded review passes

Role: editor/integrator. Task: build an edit plan from {{APPROVED_CLIPS_AND_INTERVALS}}, {{SCRIPT_AND_DELIVERY_TARGET}} and {{DIALOGUE_AMBIENCE_FOLEY_SCORE_STEMS}}.
Output: source selection with handles and rejected ranges; EDL; story/genre-appropriate transition rationale; dialogue sync and sound bridges; shot-match colour notes; editable opening/end-card text; missing coverage; targeted repair tasks with responsible roles.
Review passes: story and causality; continuity and every boundary; sound/voice sync and uninterrupted listening; technical/export/rights/rubric evidence. Assign independent reviewers where available and record self-review honestly. Repeat a pass only after a relevant change or unresolved defect, within {{ATTEMPT_AND_CREDIT_LIMITS}} and delivery reserve.
Verify: no clones, repeated action, hidden reference frames or stale source revisions; preserve approved score during picture repair. Do not force trailer cuts or action-movie rhythm onto another genre. Use observable criteria rather than promises of a named studio's quality.

P20 — Preproduction capability and feasibility plan

Role: producer with AI model/prompt lead. Task: assess {{PROJECT_ID}} against {{STORY}}, {{BUDGET_AND_DEADLINE}}, {{AVAILABLE_TOOLS}} and {{DESTINATION_REQUIREMENTS}}.
Output: phased production plan; current capability evidence for required image/video/audio/editing actions; exact account/project target; required assets; hardest-shot feasibility/fallback; requested versus verified model/duration/resolution/reference limits; cost unit and bounded spend plan; tools unavailable; gates and owners.
Verify technical claims against current official documentation and actual selected interface before their first use. Record source/date; distinguish available from authorized. Do not infer supported model controls, prompt character limits, competition rules or pricing from historical lesson text. Use the user's current model preference; make no promise that a cheaper agent matches another model or cannot hit usage limits. Return a plan, with no account reconnection or generation.

Optional department depth

Keep P01–P20 as the stage sequence. Append selected M01–M31 department modules for specialist decisions, using the coverage matrix to assign owners and outputs. Their shared envelope takes actual project facts; the existing variable dictionary remains the core prompt contract. Apply the filmmaker intent module before P04/P17/P19 when it helps clarify character, geography, light and sound. Fill the timed packet so departments work from the same beats. Do not concatenate all modules into every generation prompt.

The prompt variable dictionary

Replace every double-braced name before use. This dictionary covers all 87 variables in the prompt library. Values may be concise text or a path to pinned evidence when the receiving tool can read it. A media generator requires concrete shot text and actual attachments, not local document paths it cannot access. unknown blocks a required preflight field; none means verified absence.

Thirty-one specialist prompt modules

Append the relevant module to a core prompt or bounded task. Use the same approved brief, canon, assets and scene ID. Replace bracketed context in the shared envelope with your actual information; no default actor, genre, tool, lens, music or spend is implied. Select only modules needed for this shot. Definitions, ownership and acceptance criteria are in the department playbook.

Shared envelope

Project / scene / shot: [IDs]
Task owner / module IDs: [roles and selected Mxx modules]
Pinned inputs: [brief/script/canon/asset versions, paths and hashes]
Current stage / requested deliverable: [specific output]
Authority and limits: [allowed actions, write scope, time, cumulative attempts/spend]
Observed tool capabilities: [supported controls and unknowns]
Apply the selected specialist module below. Preserve approved canon and reference scopes.
Return the artifact/candidate version, decisions, evidence-backed acceptance results, limitations, unresolved dependencies and next owner. Unknown is not pass. Propose only the smallest necessary next step; a planning prompt is not authorization to launch media jobs or publish.

M01 — Development and story

Owners: EP / WR. Expected artifact: Brief, beat sheet, script, research/world-rule log, runtime budget.

Turn the supplied premise into an observable causal chain. For each beat name the audience information change, required image/sound and minimum coverage. Identify one costly dependency and a story-preserving simpler version.
Acceptance: Each required payoff has a visible/audible setup; runtime arithmetic includes all material; claims needing evidence are marked.
Known limitation: A fluent script is not proof of production feasibility.

M02 — Production, line production and assistant direction

Owners: PM / EP. Expected artifact: Schedule, task graph, budget/attempt ledger, daily status, fallback decision.

Build the smallest executable wave plan. State dependencies, owner, input versions, outputs, acceptance tests, deadline, remaining attempt scope and fallback. Calculate whether remaining work fits the reserve.
Acceptance: Critical path plus protected reserve fits deadline; rerouted/renamed jobs retain prior spend and attempts; uncertain jobs are reconciled.
Known limitation: Queue estimates vary; a planning document cannot enforce leases, payments or deadlines by itself.

M03 — Direction, casting and performance

Owners: DR / WR / SD. Expected artifact: Director intent, casting/voice decision, performance card, approved take selection.

Write a playable performance card: desire, concealed feeling, gaze target, posture, action, pause and end state. Supply exact words only where authorized. Describe an observable pass/fail and a quiet fallback.
Acceptance: Action communicates the intended change; expression does not contradict subtext; identity, accent and delivery stay consistent; exact dialogue is checked by listening.
Known limitation: Subtle acting and lip sync are unreliable; select takes, shorten beats or use an editorial reaction rather than piling on adjectives.

M04 — Visual development and concept design

Owners: CV / DR / PD. Expected artifact: Look bible, approved concept set, palette/texture sheet, reference authority notes.

Propose a compact visual grammar with a reason for each choice. Separate immutable design from scene-specific lighting/state. Return reference scopes, representative checks and a cheaper implementation path.
Acceptance: Across representative day/night/interior/exterior needs, motifs remain recognizable and support the story; references have explicit permitted influence.
Known limitation: A mood board may contaminate faces or geography if attached indiscriminately; expensive equipment tags do not establish a look.

M05 — Art direction, sets and environments

Owners: PD / CV / CC. Expected artifact: Location bible, plan/elevation sketch, view list, set-dressing and state map.

Map the scene location from its action requirements. Label fixed landmarks, paths, camera zones, light sources and approved views. Identify unsupported angles and the reference needed before generation.
Acceptance: A reviewer can map adjacent shots to the same place; door/window positions, usable space, landmarks and relative scale agree.
Known limitation: AI camera changes can rebuild the set; approved alternate views or a simple 3D/previs plate may be needed.

M06 — Props, mechanisms and practical action

Owners: PD / CC / FX. Expected artifact: Prop sheet, state diagram, contact/occlusion map, mechanics test and insert requirement.

Describe the object at rest, contact, movement, consequence and final state. Name the hand, contact point and occlusion. Design the shortest test that proves causality and a fallback that does not invent missing action.
Acceptance: Count and scale stay fixed; fingers contact the correct surface; release follows support; mechanism moves once and causes the intended result.
Known limitation: Text cannot reliably repair contradictory image geometry; split close interaction into supported shots, composite or simplify.

M07 — Costume, hair and makeup continuity

Owners: CC / PD / DR. Expected artifact: Character turnaround, costume detail sheet, appearance-state timeline.

Create an appearance continuity card with invariant details, authorized state changes and excluded additions. Explain which references control face, body and clothing; list checks at wide, medium and close framing.
Acceptance: Silhouette and distinctive details match at relevant framings; lighting does not disguise an unapproved redesign; changes occur only at approved story events.
Known limitation: Repeated new character generations often create a new person; post-lock work makes views/states of the approved identity.

M08 — Cinematography: composition and shot language

Owners: DP / DR / SB. Expected artifact: Camera/coverage plan, composition frames, reveal map and framing tolerances.

For each beat choose framing, camera position/height, subject placement and what remains unseen. Explain the narrative reason, required visible detail, cut partner and observable composition check.
Acceptance: The story subject/action reads at delivery size; composition carries the intended information; reframing/crops preserve necessary hands, props and gaze room.
Known limitation: A framing label is approximate in generated media; inspect actual composition, not just the prompt.

M09 — Lenses, perspective, focus and depth

Owners: DP / SB. Expected artifact: Lens/geometry intent card, focus map, near/far anchors, supported-control record.

Describe the viewpoint, distance relationships, field of view, sharp region and focus target in observable language. Add numerical lens values only with an explicit assumption/control. Give a static-focus fallback.
Acceptance: Near/far size relationships fit the intended viewpoint; critical target is readable; a focus pull changes sharpness rather than morphing objects.
Known limitation: Camera position determines perspective; focal length and sensor format affect field of view. Numeric lens/aperture prompts are intent cues unless the tool implements them.

M10 — Camera operation, blocking, movement and grip translation

Owners: DP / SB / CC. Expected artifact: Camera path, blocking/eyeline map, axis notes, motion reference and boundary card.

Plan one primary camera move around one action. Record start/end framing, path, pace, screen direction, eyeline target and axis side. Explain any crossing and provide a locked-camera coverage alternative.
Acceptance: Landmarks/parallax remain plausible; gaze has a stable target; entrances/exits preserve direction; any axis crossing is re-established visibly.
Known limitation: Grip/support names describe desired stability or path, not real hardware in AI; complex orbit plus fine interaction is high risk.

M11 — Lighting design and gaffer translation

Owners: DP / PD / CF. Expected artifact: Lighting diagram/state map, approved keyframes, shadow/reflection continuity notes.

Describe motivated sources and their effect on subject/background, including direction, softness, contrast and colour relationship. Define before/after states for practicals and checks on shadows, reflections and exposure.
Acceptance: Faces/action remain readable; shadows, reflections and highlights agree with sources; a source change affects the appropriate surfaces and adjacent shots.
Known limitation: Lighting prose is approximate; invented fixtures or inconsistent shadows require reference/selection/compositing. No physical rigging instructions are provided.

M12 — Camera pipeline: exposure, motion and colour intent

Owners: DP / CF / AR. Expected artifact: Image pipeline note, source metadata ledger, transform/LUT provenance, reference export.

Specify visible exposure and motion goals, then separately list verified file properties and supported settings. Map input→working→output colour handling and flag unknown transforms before grading.
Acceptance: Important detail survives highlights/shadows; cadence and blur suit action; transforms are applied once; output tags/spec match actual encoded content.
Known limitation: Writing cinema/log/HDR in a prompt does not create corresponding metadata or latitude. Standards workflows are optional and require correctly identified inputs.

M13 — Storyboards, animatics and previs

Owners: SB / DP / FX. Expected artifact: Numbered panels with state labels, timed animatic/shot list, source-to-timeline duration map.

Build a timed board plan with exact panel count, camera/action/state/sound and incoming/outgoing boundaries. Separate edit duration from supported generation settings. Label every board planning-only or attachable, and justify it.
Acceptance: Panel count equals stated grid; source clips cover edit duration plus handles; shot boundaries preserve axis/state; each cell shows one moment.
Known limitation: A multi-panel board may generate a collage; planning frames are not automatically valid start-image inputs. Animatic rhythm does not prove generated timing.

M14 — Script supervision and continuity

Owners: CC / ED / DR. Expected artifact: Continuity ledger, cut-boundary cards, discrepancy report, change-impact list.

Compare incoming/outgoing states across this shot and its neighbours. Return exact mismatches with source times, scope, required modality and smallest repair. Do not promote a rejected frame to a reference.
Acceptance: Both sides of each join match or have a deliberate motivated transition; accepted state comes from actual selected frames; contradictions have an owner/decision.
Known limitation: A still cannot establish temporal continuity or sound; inspect full relevant motion and listen when required.

M15 — Model feasibility and prompt engineering

Owners: ML / DP / DR. Expected artifact: Final prompt, negative/exclusion list where supported, attachment map, capability/unknown log.

Compile the packet into one model-compatible prompt. Preserve exact canon and dialogue, separate settings from prose, resolve conflicts and name the highest-risk request. Return attachment IDs and observable acceptance tests.
Acceptance: Prompt is complete and internally consistent; each reference has control/exclusion scope; critical requests map to supported settings or tested intent cues.
Known limitation: Long prompts and named premium models do not guarantee compliance; unsupported exact timing/control must remain an uncertainty.

M16 — Generation operation and job reconciliation

Owners: GO / PM / AR. Expected artifact: Preflight evidence, job ID/status, actual cost, original output and provenance record.

Prepare or inspect the preflight record for this authorized operation. Compare expected versus observed account, prompt, attachments, mode, settings and cost. Return hold on any unresolved mismatch; do not infer submission.
Acceptance: Observed composer matches expected state; job/output mapping is exact; no duplicate submission or exhausted scope; candidate status remains unapproved.
Known limitation: A UI click or progress badge is not proof of attachment, export or job success. Tool/account availability must be checked at execution.

M17 — Asset library, DIT and data management

Owners: AR / CC / PM. Expected artifact: Asset registry, checksums, source-proxy map, manifest, backup/restore evidence.

Inventory sources and candidates with versions, hashes, approvals, dependencies and intended reference scope. Identify missing/duplicate/stale entries and propose recoverable active pointers without deleting originals.
Acceptance: Every accepted artifact resolves to exact bytes and provenance; proxies map to originals; a sample restore/relink succeeds within the declared scope.
Known limitation: Hash equality proves bytes, not creative approval or rights. Avoid leaking private evidence into reusable/public packages.

M18 — VFX supervision, compositing and repair

Owners: FX / DP / CC. Expected artifact: VFX shot brief, layer/matte package, before/after, change mask, reviewed composite.

Design the smallest repair with exact frame range, protected elements, required layers, tracking/contact/lighting rules and acceptance checks. State what cannot be reconstructed honestly from the available source.
Acceptance: Effect attaches to the world with correct scale, occlusion and timing; protected identity/background remain unchanged; no matte chatter, halos or invented action.
Known limitation: Video inpainting may regenerate outside the mask; unavailable depth/clean plates limit repair. Compare all affected frames, not one hero frame.

M19 — Editorial, assistant edit and conform

Owners: ED / AR / DR. Expected artifact: Edit project, EDL/source-time map, review cut, picture-lock record, change list.

Build an edit decision plan with source in/out, timeline in/out, story purpose, transition and sound bridge. Flag missing coverage. Preserve approved music timing during picture-only repair unless a change is authorized.
Acceptance: Every timeline event resolves to an approved source; action/consequence remains visible; all joins/cards/crops pass; reconform retains timing and audio sync.
Known limitation: Editing cannot manufacture missing causality; generative text should normally be replaced by controlled typography. New picture timing invalidates affected audio/reviews.

M20 — Sound supervision and spotting

Owners: SD / DR / ED. Expected artifact: Spotting/cue sheet, sound bible, cue hierarchy, stem plan, sound risk list.

Spot this cut beat by beat. Name the audible focus, listener position, required cues, silence, source/rights status, sync point and edit tails. Identify contaminated or unavailable stems explicitly.
Acceptance: Each story-critical audible event has a source/owner/time; dominant cues are identified; picture changes propagate to relevant cues.
Known limitation: A native mixed soundtrack may be inseparable; an audio stream does not prove dialogue, sync or clean stems.

M21 — Dialogue recording, voice, ADR and dialogue edit

Owners: SD / DR / LR. Expected artifact: Dialogue takes and selects, pronunciation sheet, sync map, clean dialogue stem, transcript.

Prepare a voice/ADR card with exact text, pronunciation, objective, pauses, breaths, timing window and approved voice source. Define sync anchors and an editorial fallback if speech cannot fit naturally.
Acceptance: Every word matches approved text; voice identity/perspective is consistent; no clipped consonants or obvious sync offset; line remains intelligible in full mix.
Known limitation: Cloned likeness/voice requires applicable permission; plausible timbre is not proof of consent. Lip sync and denoising artifacts need actual listening/viewing.

M22 — Foley performance and edit

Owners: SD / ED. Expected artifact: Foley cue sheet, separate clips/takes, sync notes, Foley stem.

List only meaningful visible contacts with time, material, weight, perspective and dynamic priority. Write short individual sound briefs, then specify sync/trim checks and where Foley should remain understated.
Acceptance: Contacts occur at visible impact; surface, weight and distance fit; footsteps do not continue after stopping; no needless emphasis masks performance.
Known limitation: Sound generation may produce extra impacts or wrong materials; split cues, trim and sync manually rather than accepting a whole noisy bed.

M23 — Designed sound effects

Owners: SD / FX / DR. Expected artifact: SFX design sheet, licensed/generated sources, layered session, rendered effects stem.

Design the effect from its physical or fictional cause. Specify onset, body, tail, material, distance and emotional role. Separate literal mechanism from optional dramatic layer; give sync and contamination checks.
Acceptance: Sound follows the intended mechanism once; onset/decay and space match picture; layers avoid unintended associations or masking the causal cue.
Known limitation: A cinematic whoosh is not a substitute for a missing switch/contact; generated effects may contain unwanted speech or music.

M24 — Ambience, room tone and acoustic continuity

Owners: SD / ED. Expected artifact: Ambience map, clean beds/loops, room-tone patches where relevant, edit tails/stem.

Map each location to a restrained base bed and optional events. Specify perspective, transition overlap, loop-safe ranges, exclusions and how the bed changes with the character crossing a boundary.
Acceptance: Cuts do not expose holes or abrupt noise changes; loops avoid audible repetition; distance/interior/exterior shifts are motivated; beds leave speech room.
Known limitation: Static images cannot verify acoustic continuity; generated ambience may add anachronistic traffic, voices or score.

M25 — Music, composition and clearance

Owners: SD / DR / LR. Expected artifact: Music brief, cue map, source/license log, final masters/stems if available, temp-replacement list.

Write a music brief from the scene emotion and structure: cue purpose, motif, instrumentation, start/release/silence and protected timing. Mark temp/final status, rights evidence and missing stems; do not add score by default.
Acceptance: Every final cue has an approved source/use scope; temp tracks are replaced or cleared; entrances/releases serve story; critical speech is not masked.
Known limitation: A platform subscription alone does not establish every distribution right; generated music may lack clean stems. No filmmaker/composer imitation is required.

M26 — Sound edit, re-recording mix and mastering

Owners: SD / ED / QC. Expected artifact: Mix session, final mix, aligned D/M/E stems where available, measurement report, listening/QC notes.

Create a mix plan around the dominant cue per beat. Specify balance, automation, perspective, transitions, stem alignment and measured destination targets. Separate measured compliance from listening judgments and list unavailable checks.
Acceptance: Full mix is intelligible and free of clipping/clicks/gaps; sync holds start to end; measured target matches destination; recombined stems match the intended mix or differences are documented.
Known limitation: There is no universal web loudness target. A limiter cannot restore clipped source; one headphone/device check does not certify every playback environment.

M27 — Colour, online and finishing

Owners: CF / DP / ED. Expected artifact: Conformed timeline, grade/project, graphics sources, export settings, versioned review master.

Write the conform/grade/export plan from actual metadata. Separate balancing from creative look, protect approved timing and list comparison frames plus full-motion checks for artifacts and graphics.
Acceptance: No stale/proxy source remains unexpectedly; neighbouring shots match intent; no crushed essential detail, banding or unintended colour shift; mix/timing/captions match lock.
Known limitation: Grading cannot recover absent information or fix physical continuity; a LUT is not a substitute for correct source identification.

M28 — Quality control and accessibility

Owners: QC / SD / LR / DA. Expected artifact: Timecoded QC report, modality coverage, corrected access files, pass/hold per requirement.

Audit the supplied exported artifact against explicit requirements. Record timecoded findings, what you viewed/heard/measured, accessibility checks and unknowns. Sweep similar boundaries after a repeated defect; report release hold only for actual unmet requirements.
Acceptance: Every required modality is actually reviewed; caption text/timing survives delivered playback; required failure/unknown cannot be averaged into pass.
Known limitation: Automated transcription/metrics and sparse contact sheets cannot certify full viewing/listening. Applicable access/destination needs must be established, not invented.

M29 — Rights, provenance and release review

Owners: LR / EP / AR. Expected artifact: Clearance/provenance matrix, attribution/disclosure list, specific release holds.

Reconcile the final asset list with evidence of use scope and current destination requirements. Return supported, missing or disputed facts with exact sources and smallest resolution; do not infer permission from availability.
Acceptance: Every included external asset/likeness/voice has applicable evidence or explicit unresolved status; current requirements are verified at release time.
Known limitation: This is an evidence workflow, not legal advice or a claim that generated material is automatically clear. Seek qualified counsel for material legal uncertainty.

M30 — Delivery, distribution, archive and handover

Owners: DA / AR / PM. Expected artifact: Delivery manifest, playback evidence, submission receipt, archive manifest, handover/lessons.

Prepare a destination-specific release and archive receipt. Separate local export, upload, published playback, submission and recovery evidence. List missing checks and next owner without declaring completion from a status badge.
Acceptance: Uploaded version matches approved content; required link/visibility/captions work; actual submission is confirmed; archive omissions and restore test are explicit.
Known limitation: Upload success is not submission success; a public link can change later. Archive manifest alone does not prove restore without testing.

M31 — Physical production, stunts and specialist applicability

Owners: PM / DR / LR. Expected artifact: Applicability record and human specialist handoff; synthetic staging constraints.

Classify planned work as synthetic, ordinary real recording or requiring specialist review. Name relevant human expertise and unresolved dependencies. For purely AI scenes give continuity/visual intent only; do not generate real hazardous execution instructions.
Acceptance: Every physical activity has an appropriately qualified responsible human and required local process; AI artifacts remain creative planning only.
Known limitation: Prompts do not replace on-set expertise, inspections or real safety planning; this package provides no stunt/rigging/hazard execution procedure.

Project brief and canon template

Status: DRAFT. A filled form is not approval. Replace brackets; record unknowns explicitly.

Brief

Project ID / title: [ID / title]
Owner and director: [named humans; delegated limits]
Audience / destination / emotional promise: [specific viewer and intended experience]
Premise / protagonist intent / obstacle / consequential choice: [story]
Beginning → change → ending: [observable progression]
Runtime including cards: [seconds]; picture FPS / dimensions / ratio: [verified target]
Audio / captions / access needs: [target and dated source]
Style and exclusions: [observable aesthetic choices]
Budget: [amount and unit]; deadline with timezone: [timestamp]
Protected edit / QA / upload-transcode reserve: [hours, rationale and uncertainty]
Picture-lock cutoff / pivot checkpoint: [timestamps]
Maximum cumulative shot attempts / spend: [limits, including failed jobs]
Approved fallback: [coherent alternate coverage or smaller story; approval pointer]
Binding references / inspirational references: [separate lists]
Rights / consent / destination requirements: [evidence or unresolved item]

Canon v001

Camera axis / anatomical versus screen direction: [map]
Sound world / voice / score policy: [rules]
Story-state lighting progression: [states]
Known failures: [specific negatives, applicable shots]

Change control

Editorial adjustment: trim/grade inside approved meaning and boundaries; log it and recheck joins.
Material change: identity, hero-prop geometry/count, geography, causal action, dialogue meaning, approved voice, runtime/delivery mandate. Propose a new version and dependency impact; human DR/EP decides before promotion.

Change ID / old and proposed versions / reason / affected descendants: [record]
Cost and critical-path impact / fallback / decision evidence: [record]
Do not reset attempt counters by renaming a shot.

The timed scene packet

Use only relevant department modules; assign an owner for each selected module. A packet records decisions, not permission to generate. Bracketed text below is replaced with project facts; unknowns stay explicit.

Intent, canon and authority

Project / scene / shot / clip IDs: [stable IDs]
Brief / script / canon versions: [pinned revisions and hashes]
Selected modules / owners: [Mxx → named role/actor; relevant omissions and reasons]
Director intent: [desire, obstacle, subtext, choice and visible emotional change]
Audience information: [what is known before, what is revealed, what remains concealed]
Story purpose / beginning / change / ending: [observable]
Required subjects / excluded subjects: [lists]
Incoming approved state and actual selected boundary: [asset, source time, frame, pose, gaze, hand, prop state, camera side, light, sound tail]
Outgoing required state: [same fields]

Visual and physical plan

Location geography: [entrances/exits, landmarks, near/far anchors, actor paths]
Camera: [shot scale, viewpoint/height/distance, composition, field of view, optional supported lens intent, focus target/depth, camera path/pace]
Continuity: [axis side, screen direction, eyeline target, intentional re-establishment if crossing]
Lighting: [motivating sources, direction/softness, fill/contrast, colour relationship, practical state, shadows/reflections]
Appearance/props: [approved face/body/costume/hair/makeup and prop count/scale/mechanism; allowed changes]
Image pipeline: [actual fps/cadence, motion-blur intent, exposure priorities, verified source/working/output colour handling]
Tool feasibility: [supported settings versus prose intent; difficult contact/occlusion; smallest test and editorial fallback]

Final runtime sums retained timeline intervals, transitions and cards. Budget unused source handles separately; three six-second sources retained at four seconds each contribute twelve seconds, not eighteen.

Shared timed beat plan

Use one timing basis, [scene-relative seconds / timeline timecode and fps], across both tables. End times are exclusive. Record source-time separately after selecting footage. Render length is a verified tool setting; desired edit time is not proof that it is supported. Every beat ID appears in both tables, even when the correct audio choice is silence.

Audio sources / voice identity / pronunciation: [approved IDs, rights status, temp/final]
Stem and sync plan: [dialogue/music/effects or finer stems available, alignment, tails, missing clean sources]
Edit change policy: [picture-only versus timing change; affected score/ADR/captions/reviews]

References, acceptance and handoff

Prompt: use P05 with selected specialist prompt modules. Fixed canon text must match the approved state; variable block describes this shot only.
Acceptance test IDs: [5–8 critical observable criteria, plus relevant delivery requirements; required modality and reviewer for each]
Cost / model / mode / account / project / batch / settings: [pending until live readback]
Authorization ID and scope / attempts remaining / deadline: [evidence]
Preflight result / job ID / actual cost / submission status: [observed only]
Review result / retained interval / boundary / next owner: [pending until reviewed]
Required output package: [packet, exact prompt/attachments, source/job, selected interval, review, boundary, sound cues/stems, changes/unknowns]

Clip review and handoff form

Reviewer identity / role / timestamp: [record]
Independent of maker: [yes/no; self-review is valid when labelled]
Candidate ID/version/hash / canon / packet / neighbours: [pins]
Review coverage: full motion [yes/no]; audio listened [yes/no]; frame sampling [exact times]; metadata [checks].
Rubric name/version/source: [provisional workflow rubric or verified named source]

Hard gates: story outcome, identity/props/geography, readable physics, start/end state, picture/audio, delivery requirements. Every applicable hard gate must pass.
Ratings: story / continuity / performance / camera-editability / sound / technical, each 0–4 with evidence. Never average away a hard failure.
Usable source interval / timeline location: [start/end in seconds, end exclusive]
Outgoing boundary: [actual frame and complete state]
Verdict: [candidate usable / revision required / rejected / unverified]; creative acceptance remains a separate DR decision.

Targeted revision request

Finding ID / severity / failing interval: [record]
Expected / observed / likely cause / uncertainty: [record]
Smallest repair / owner / pinned inputs / test to close finding: [record]
Affected neighbours and downstream artifacts: [list]
Other shots sharing this root cause checked: [scope/result]
Remaining cumulative time / attempts / cost / fallback: [record]

Worked example: The Last Parcel

This is a synthetic rehearsal of the workflow, not a produced film. Names, timings and limits illustrate decisions; they are not defaults. Use blank forms for a real film.

Mandate and locks

Human Mira is EP/DR. Producer agent P coordinates; writer W, visual maker V, editor E and reviewer Q are distinct actors. Q never edits a candidate it certifies. Mira can combine specialist roles in a solo run; label resulting self-review honestly.

Brief LP v1: a courier returns a repaired toy in a parcel to a workshop. The ending reveals who repaired it through an approved written note added in the editor. No spoken dialogue. Target edit 24 seconds including a 2-second end card. Overcast dawn, restrained natural light, one workshop. Three shots: arrival 8 seconds; placement 6; empty doorway and parcel/note insert 8. A camera change inside the last beat means separate source clips; edit duration stays 8 seconds total.

Canon LP v1: blue coat; one rectangular parcel; one red cord; parcel carried in right hand; left-hinged workshop door; same bench and door geography. Note text is human-approved editable typography, not generated writing. Reference IDs CHAR01, PROP01, LOC01 are synthetic labels only. A real run replaces them with inspected IDs and files.

Example authorized ceiling: 3 attempts and 120 mock credits per shot; 300 mock credits total; 6 hours until deadline, including a protected 90-minute edit/QA/upload reserve. No actual approval or spend occurs here.

Producer dispatch and integration

All pass/fail events above are stipulated example outcomes. Offline fixtures exercise the data contract separately; they do not establish any real picture or sound quality.

Example change, conflict and deadline decisions

After placement is accepted, visual maker V proposes a round parcel because it looks better. CC identifies placement, note insert, continuity ledger and current edit as descendants. AR does not promote it. Mira retains rectangular canon v1; no job launches. If Mira had approved canon v2, PM would invalidate only affected tasks/results and assign new reviews before integration. Sound not dependent on parcel geometry could retain its evidence with a scope note; contact Foley would need review.

If another worker has used red string and another blue ribbon, the producer does not blend prompts. AR presents reference/version facts, CC reports the mismatch, and DR selects the approved design. The losing candidate stays recoverable.

Suppose 3 hours remain. The remaining generation critical path is estimated at 2 hours, edit/QA/upload needs the reserved 1.5 hours: 3.5 > 3, so more generation does not fit. PM freezes optional polish. DR proposes already-approved placement footage plus an existing static note insert and sound bridge. If that preserves causality and fits 1.5 hours, EP approves the fallback. If the required placement action is absent, this fallback fails; EP chooses a smaller newly briefed film or defer/abandon. The deadline never converts an unseen action into a completed one.

Filled department packet: parcel placement

This is the proposed six-second usable placement shot within the 24-second rehearsal, not a generated result or a claim that every tool offers a six-second setting. Preflight selects a supported render length with handles. Timing below is clip-relative, end-exclusive; in this example it begins at timeline 8.0 seconds. Source offsets are entered after a real take is selected.

Director intent: the courier chooses to leave the repaired toy without asking for recognition. Show deliberate release, not a triumphant smile. Geography: the same bench beside the left-hinged door; retain the arrival shot's axis. Camera: locked close view from the accepted side, enough width for wrist, parcel and support surface; focus stays on contact. Soft overcast dawn remains stable. Numerical lens labels add nothing unless they help reproduce an actual supported framing.

Selected modules: M03 performance; M06 prop mechanics; M07 coat/hand appearance; M08/M09 framing/focus; M11 light; M14 continuity; M19 edit; M20/M22/M24/M26 sound. Existing blue coat, rectangular parcel and one red cord remain fixed. Music is not required for this illustrative shot; no dialogue is added.

Acceptance: one parcel/cord, same scale and right hand; support precedes release; no floating/intersection; stable geography/light; usable boundaries match neighbours; one contact sound on actual impact with no extra music/voice. Full relevant motion and listening are required. If the fingers fail, return a bounded M06/M15 repair; if a camera move hides contact, return to M08 rather than changing the story. ED cannot trim away the only visible proof of placement. These are expected criteria, not observed passes.

Handoff receipt

The final archive would contain brief/canon/script versions, inspected source registry, shot graph, prompts and real job records, accepted intervals/boundaries, edit project/EDL, audio stems, note/title sources, master/checksum, review evidence, rights records and submission receipt. This rehearsal returns a design and expected decisions only. A real creator fills the forms with observed evidence as each gate is reached.

Instructor sequence: retained and improved

Source: complete supplied text, SHA-256 [source hash retained privately]. Read in full on 17 September 2026. The first line requests a Night Watch customization; that adaptation is separate from the generic library. Remaining paragraphs are historical evolving production instructions, treated as source evidence. They do not authorize any action in this task.

The primary method remains the user's eleven-stage handbook. The Shameer notes supplement audience promise, spoken timing and destination reframing. This newly supplied sequence adds concrete role prompts, asset experiments and the staged-composer approval pattern.

Entire sequence mapped in source order

Conflicts resolved explicitly

1. Storyboard count: 7–8 frames is not a 4×4 grid. Use exact panel count; 8 can use 4×2, 16 uses 4×4. Panel ratio and sheet ratio are independent. More panels do not require more generated shots.
2. Model and duration: the sequence changes from a draft model to direct final-model generation. This is a historical revision, not two simultaneous rules. Confirm current supported settings and the latest user choice. Scene duration never proves a particular render setting is available.
3. Runtime arithmetic: 15 + 20 + 30 = 65 seconds. A 75-second target needs 10 more retained seconds; an 80-second target needs 15. Count cards and transitions explicitly. Do not add a fourth scene merely to preserve mistaken arithmetic or invent coverage to fill time.
4. Method 2: the first detailed definition creates a board but leaves it out of the generation attachments; a later summary says it is attached. For this generic experiment define B as planning-only board. Label that resolution as our chosen experiment definition and seek a real decision before executing a differently intended test.
5. Method 4: colour board alone is D. The later request adds character sheets, creating D2. Neither is a recommended default until tested. A whole board may cause collage/reference-flash artifacts; using individual approved frames is a separate candidate method.
6. Options versus locked identity: explore new design options only before lock; after lock, create views or state variants of the same identity. Three options and batch location generation are optional policies, not mandatory spend.
7. Length versus clarity: preserve emotional orientation, references, exact action/dialogue and handoff; trim irrelevant adjectives. Verify any character limit instead of filling a claimed 10k–20k allowance.
8. Review and authority: the historical sequence's final request is not current permission to edit or generate. This task designs prompts and procedures. Its action tone, cast, scene recreation, account reconnection and model-switch demands do not become defaults.

Prompt chain and source fidelity

For a new film: P20 → P01 → P15/P02/P03 → P16/P17 → P04/P14/P05 → P06 → P07 → P08 where needed → P19/P09 → P10. P12 coordinates ready work and P11 handles drift/deadline decisions. P18 is optional experimentation, not a required production stage.

The prompts are newly authored synthesis, not verbatim instructor templates. They preserve stage techniques while replacing story/cast/place/prop/model/duration/aesthetic facts with named variables. Every prompt names a role, required inputs, return shape and verification. Paid generation, major canon change, final cut and submission retain human gates. No inference here identifies the missing Divinium rubric.

Shantanu masterclass: story and direction controls

Three slides from the 26 August masterclass provide a compact creative-control method. Use the controls below in every project; they supplement the detailed workflow and checklist.

1. Build a story engine before generating shots

For character-driven narrative work, record five linked decisions:

1. Character — whose experience gives the audience someone to follow?
2. Desire — what concrete outcome do they want?
3. Obstacle — what makes that outcome difficult or costly?
4. Conflict — what decision, confrontation, or sacrifice must occur?
5. Transformation — what has changed by the ending?

This is the story test for a character-driven treatment, script, scene map, and final cut. Informational, abstract and sensory films may instead define an approved progression of attention, information or feeling. Record that alternative explicitly rather than forcing a character arc where it does not belong.

2. Choose one emotional direction for a sequence

Write the intended audience feeling, then align six levers around it:

The rule is alignment: do not ask camera, light, colour, sound, edit, and impact to make competing emotional promises unless that contradiction is intentional and recorded.

3. Write a director’s brief before asking an AI to generate

Use this five-part structure for prompt packets, tool prompts, review tasks, and team handoffs:

# Role
Who is acting and what relevant expertise should guide the response?

# Objective
What single result must be delivered?

# Context
What story, canon, audience, source material, constraints, and current state matter?

# Instructions
What ordered actions, output format, and quality checks are required?

# Notes
What constraints, edge cases, exclusions, formatting rules, or exceptions remain?

Keep the goal singular, state background that changes the decision, give one action per instruction where possible, and put exceptions where they cannot be overlooked.

Where the controls enter the studio

These are creative planning controls, not automatic quality certification. Record deliberate departures in the project notes.

Night Watch: existing-film adaptation

Planning/reference adaptation only, requested with the instructor sequence. The submitted film is preserved. These prompts do not authorize new images, video, uploads, account changes or edits. Use this as a worked project application and recovery guide; a later production request would need its own current inputs and authority.

Grounded baseline

The working pack defines an older watchman in a seaside amusement park before its demolition, following his daughter's remembered order of rides. The daughter is memory, not a present-day second cast member. The cold open loses a brass key through a grate, followed by SIX HOURS EARLIER. The carousel has one stubborn bulb; he climbs, unscrews and pockets it. The Ferris runs and darkens top-down. He sits by the water and addresses memory. The method plan specifies the intended sole line: “I left the lights the way you liked them.”

Locked design direction: weathered older man, grey hair, dark work jacket, boots and flashlight; cool moonlight against residual sodium/carnival tungsten; camera follows or waits, with a lead permitted at the Ferris. Native sound is predominantly wind, waves, wood, metal and clicks; a continuous cello stem belongs in post. Existing source plans differ on whether the sole line is native or added later: preserve the accepted actual soundtrack, and treat new voice work as a separate decision.

The older plan's 3:27 is a planned story runtime. The identified revised submitted master is 216.859977 seconds (3:36.86), 1920×1080, 30fps, H.264/AAC, SHA-256 [source hash retained privately]. Do not force it back to 3:27 or an instructor's 60–80 seconds. Historical submission confirmation is recorded in “Get CapCut help”; current live platform status and a fresh full listening pass are not asserted here.

The original four-person team instruction was superseded by the user's solo Night Watch entry. Historical rules, platform-only generation/enhancement claims, model choices and deadlines remain dated project records, not current universal requirements. This adaptation preserves artistic canon while leaving actual supported settings and external requirements to current verification if work is later authorized.

Asset scope to carry into every relevant task

These are recovered labels, not newly verified live asset handles. Resolve the selected asset and actual platform ID before use. The original “every shot” attachment instruction is narrowed by relevance: a hand/key ECU should not receive an unrelated whole park video; pure scenery needs no face-conditioning merely because the file exists. The approved image controls key shape; text must not demand a conflicting cutout. No snake, martial-arts cast, tournament, mentor or action-trailer tone transfers from the instructor example.

Copy-and-paste stage prompts

NW01 — Recover and plan; producer / asset librarian

Audit The Night Watch read-only. Use the local VR_Higgs working pack, method plan, scene inventory, review/revision_v2 records and latest revised master. Reconcile planned versus generated versus edited versus submitted states. Return an inventory of all seven scenes, eight reference labels, accepted sources/intervals, current edit/master, historical approvals, unresolved conflicts and evidence limits. Preserve the revised master whose SHA-256 is [source hash retained privately]. S1A's action was included in S1B; do not duplicate it as another clip. Do not restart asset generation. Produce a recovery plan with owners and dependencies only.

NW02 — Canon and references; director / continuity / librarian

Prepare a Night Watch canon and reference-scope table from the recovered approved sources. Preserve one older watchman, grey hair, dark work jacket and flashlight; the daughter remains memory. Lock gate/carousel/Ferris/boardwalk geography, the same brass key, one stubborn carousel bulb and the actual accepted prop states. Preserve cool moonlight with dying warm practical lights, follow-or-wait camera language except the Ferris lead, and the quiet emotional progression. Resolve NW_01_FACE, NW_02_BODY, NW_08_FACE_NIGHT, NW_07_KEY, NW_03_LOT, NW_04_CAROUSEL, NW_05_FERRIS and NW_06_BOARDWALK to real files and versions. Return contradictions and missing views; propose variants of the same identity only when needed. Do not create three new actors or replace accepted assets. Keep unknown live IDs unknown.

NW03 — Performance, voice and sound; director / sound lead

Create a sound/performance bible for The Night Watch. His behaviour is deliberate, restrained and personal: routine work carries grief without an invented smile or extra dialogue. Preserve only the intended S6 line, “I left the lights the way you liked them.” Use the accepted soundtrack as authority for the actual performance; the prompt alone does not prove what was spoken. Map coastal wind/waves, gate latch, footsteps, switch clicks, wood/metal movement, bulb handling and quiet pauses. Keep the approved continuous cello stem coherent and thinning toward silence. Return cue IDs, timing anchors, source/permission pointers, native-versus-post provenance and listening checks. Do not generate speech/music or change the final mix.

NW04 — Storyboard and shot graph; cinematographer / storyboard / editor

Make a written reference-grounded storyboard reconstruction for all seven Night Watch scenes, using accepted footage and the current edit where available. Number each panel, give source and timeline in/out, camera side, framing, action, prop/light state, sound cue and next-state dependency. The seven scenes are: key loss; gate entry; carousel shutdown leaving one bulb; physical bulb removal; Ferris run and shutdown; boardwalk and one line to memory; dark wheel and closing. Separate planned timings from the current 216.859977-second delivered master, including titles and ending material. Use explicit panel count and grid; 8 panels can use 4×2, not 4×4. Preserve the 16:9 source composition and the planned approximately 74.4% safe vertical band for any 2.39 matte; do not apply a new crop. Return missing evidence, not a new film or a claim that every action was freshly viewed.

NW05 — Shot packet and reference method; prompt lead / continuity

Prepare a draft Night Watch shot packet for the chosen existing shot. Include emotional situation, exact approved asset authority, incoming state at the intended cut, one primary physical action, fixed camera setup, physical action time versus supported render length versus retained interval, native sound and score policy, exclusions and observable acceptance tests. Boards are planning artifacts unless the chosen supported generation mode and test evidence justify attachment. Do not assign four experimental methods to four different critical scenes as though this proves which is best. Preserve the current user's model choice; leave live generation settings and cost unverified until actually inspected. Return the complete prompt and attachment plan only.

NW06 — Staged UI review; generation operator / producer

Prepare a preflight checklist for the approved Night Watch packet without operating the account. It must compare expected versus observed account/project, exact model/mode, full prompt including final line, selected reference IDs/thumbnails/count, duration, ratio, resolution, audio, batch, displayed cost and cumulative attempts. Historical missing chips and prompt truncation are regression cases: stop if either recurs. A project asset's existence does not prove attachment. Show the complete staged configuration to the human before any paid action unless an existing bounded approval already covers that exact unchanged scope. This task is checklist preparation only; do not open/reconnect accounts, upload or click Generate.

NW07 — Review, repair and handoff; independent QC / continuity

Review supplied Night Watch media only to the extent it is accessible. State full-motion, frame-sampling, listening and metadata coverage separately. Check key geometry and airborne-to-impact state, all similar reference-flash joins, one bulb rather than clones, physical unscrew/pocket action, Ferris switch-on versus switch-off, top-down extinction, bench pose/leg/eyeline continuity and the sole approved line. Use timecodes and exact source versions. No score overrides failed story/continuity gates. Return targeted trim/insert/repair recommendations, cumulative limit checks, accepted usable interval candidates and next boundary state; creative acceptance remains the human director's decision. Do not declare old missing-insert notes to be current defects without checking the revised master. Divinium remains unavailable; label the operational rubric provisional.

NW08 — Edit, finishing, delivery and lessons; editor / sound / QC / archive

Prepare a read-only edit and archive assessment of the revised Night Watch master. Preserve its quiet dramatic rhythm and original accepted score; do not impose action-trailer cuts or the instructor's runtime. Reconcile source intervals, edits, cards, crop/watermark/ending material and the exact master-to-submission mapping. Distinguish shot-level review from whole-file playback and historical submission from current live verification. Return an EDL/relink gaps list, sound/color review plan, criterion-based remaining checks, archive manifest and evidence-backed postmortem. Repeat review only for relevant changes or open defects. No media edits, exports, public uploads or submission changes are authorized.

Two detailed production-card examples — drafts, no generation

S1-C: key impact and loss

Role and dependency: ML prepares; CC checks the selected S1-B airborne boundary; human DR approves any future action. The existing review records an extracted frame at S1-B source 6 seconds. This is a historical candidate boundary, not an instruction to change the submitted cut.

PRODUCTION BRIEF
The key's irreversible disappearance opens a quiet story of a watchman's final night. The wet drain grate gives the loss a clear physical destination. The approved airborne boundary controls key shape, position, scale, wet asphalt, grate geometry, camera axis and light. Do not redesign the key to satisfy contradictory prose. No whole-park video is a direct-motion reference for this ECU.

REFERENCES
Use only the inspected selected airborne boundary and any specifically required approved key reference. Record their real IDs and authority outside this generator prompt; never invent tags.

TIMED ACTION — proposed 5-second production unit, only if later verified and authorized
00:00–00:02 — Locked extreme close-up, same axis. The already-airborne brass key completes its short descent, contacts a grate edge once, tips through the existing opening and disappears below. Normal gravity, no shrinking, morphing or hovering. Keep contact readable.
00:02–00:05 — Hold on the same empty grate. No hand returns. One restrained metallic impact decays into coastal air; no music or speech.

END / EXCLUDE
Stable empty-grate state for the editor's cut to black. No balancing upright, repeated bounce, wide park view, new cutout, new angle, fade, captions or generated title. SIX HOURS EARLIER belongs in the edit.

The historical correction used a longer supported render setting with action in the first two seconds and an empty-grate hold; this proposed five-second card is a planning example, not a claim about current model controls or the final film's source duration.

S6-B: one line to memory

PRODUCTION BRIEF
After extinguishing the rides, the watchman sits facing the sea. He speaks to his daughter's memory; no one appears beside him. Keep restraint and the existing seated performance. The approved boardwalk boundary controls bench, rails, hand/leg placement and eyeline; night-face/body plates preserve the same man and jacket.

REFERENCES / STAGING
Inspect the accepted seated boundary with NW_06_BOARDWALK and relevant NW_08_FACE_NIGHT/NW_02_BODY identity scope. Do not use the unrelated old park-look video. Side view, camera waits or makes the approved restrained push; do not lead him or cut to another speaker.

BEATS — timing must be conformed to the accepted performance
Opening handle: seated, gaze toward dark water, quiet breathing.
Dialogue beat: “I left the lights the way you liked them.” One quiet line, no paraphrase or added response. Set exact start/end from measured approved audio before production.
Ending handle: retain pose and gaze, allow the line to settle into waves and wood ambience. Keep the approved cello stem separate and continuous in post.

EXCLUDE / CHECK
No second person, smile that changes the meaning, extra dialogue, reversed seated pose, new bench, abrupt camera reset or independently generated cello. Review actual spoken words and sync by listening; a text prompt does not certify audio.

Scope and evidence check

All seven scenes and eight asset labels are represented; no instructor cast or settings become Night Watch defaults. The sole intended dialogue matches the method plan and S6B saved-prompt review. Master metadata/hash was directly rechecked; 23 frames at 0, 10, …, 210 and 216 seconds were inspected. They show the key/grate, time card, watchman/gate/carousel/Ferris/boardwalk, closing and ending material. They do not establish complete motion, exact spoken words or every cut. See evidence for the historical repair/submission sources and remaining limits.

Department depth for this existing film

Use the 31-area playbook selectively with the existing canon and the Cameron/Spielberg map. This adds read-only planning guidance, not a new edit request or a claim that the latest master fails these checks. Timing is taken from the actual selected source/timeline when future work is authorized; the historical 3:27 plan must not overwrite the revised master's measured duration.

For a future scene card, copy the timed packet and use the same beat IDs for camera, lighting, action, performance, dialogue, Foley/SFX, ambience, music and edit. Fill exact ranges only from a proposed approved plan or observed source. The Last Parcel example demonstrates the form without inventing Night Watch measurements.

What we have done in the Higgs AI Film Festival

This is a case study of production controls drawn from the Higgsfield team material in the connected Drive and from project records. It is not a generic festival rulebook, a claim that every approach is required, or a current statement of any festival's terms.

Observed working patterns worth reusing

What stays project-specific

The Drive material includes particular characters, a 3:34 master, named generation routes, resolution choices, audio-off tests, a provisional ending candidate and festival compliance notes dated August 2026. Those facts belong to that project. They do not prescribe a runtime, tool, model, price, available setting, sound rule or compliance requirement for a different film.

For a festival or platform delivery, obtain the current rules directly from that destination at project intake. Add only the applicable requirements to the delivery stage and retain the current source with its date. Private handover and public release use different checks.

How this case study changes the studio

The creator path incorporates the reusable controls in stages 3 through 10: binding reference scope, one-action shots, controlled tests, typed rejection reasons, continuity handoffs, cue sheets, actual-file review and an archive manifest. The detailed workflow preserves the full gate and evidence model for teams that need it.

Source boundary

This summary is based on the connected Drive folder Higgsfield_Festival_Production_Bible_v1.0, including its start guide, credit protocol, reference plan, continuity handoff, rejection matrix, generation-log template, master-shot router, timeline cue sheet and sound cue sheet. The source files are treated as project records. They have not been redistributed or presented as official platform documentation.

Higgsfield team films: public reference list

The following links were supplied by the AIC Filmmakers community as public team-film references. They are included so creators can watch completed work and discuss visible choices. They do not reveal private prompts, source media, budgets, team roles, licenses, rejected takes, or the actual production process behind any film.

Public team films

• Team 1 — community member’s film
• Team 2
• Team 3
• Team 4
• Team 5
• Team 6
• Team 7
• Praful ji

How to use this reference set

Watch one film with a specific learning question rather than treating it as a template to copy:

1. What is the opening promise and what changes by the end?
2. Which image, sound, pacing, or edit choice makes that intention understandable?
3. Where do you see continuity, a deliberate ellipsis, or a controlled transition?
4. What would you need to know before adopting a similar approach in your own film?

Record observations as inspiration. Verify tools, rights, requirements, and source evidence independently for your own project. A public finished film is not proof of the method, permissions, or platform settings used to create it.

TypeSafe decision map

Use TypeSafe's Jev model for short, reviewable judgments in the future hosted studio. Keep the API key server-side; this local review edition makes no TypeSafe calls.

Creator guidance

Review boundaries

• Keep the existing deterministic tracker as the source of truth for progress counts, ownership, evidence, and exports.
• Treat Jev output as a suggestion with probabilities, never as proof of completion, legal clearance, rights clearance, publication authority, or a financial decision.
• Route low-confidence results to the creator instead of guessing.
• Keep question definitions and thresholds in one reviewable server-side file.
• Do not expose the API key in browser code, HTML, PDFs, exported project ZIPs, or generated prompts.

First implementation

Start with the project-brief classifier. It has low consequence: it suggests the first guided stage and a draft prompt, while the creator remains in control. Add evidence relevance only after representative project examples show that its labels are useful.

Independent review and release scope

An independent review agent assessed the workflow and checklist against solo and coordinated AI-production scenarios. It found the overall design sound and identified six corrections, all incorporated in this edition.

1. Failed and cancelled jobs have an honest terminal record without an invented output.
2. Empty, abstract, stylized and deliberately silent work have explicit applicability rules.
3. Local/private handover is separate from publication or festival submission.
4. Checks attach to a unit, revision and input/range context; changed joins or sound can stale evidence without changing source bytes.
5. Unused source handles stay outside final-runtime arithmetic.
6. Planned offscreen action, ellipsis and stylized physics are valid where consistent with the approved film.

A second review exercised the HTML tracker. It found and corrected lost draft edits, stale-tab overwrites, tracker data appearing in browser print, and malformed URL fragments. The corrected paths passed isolated browser checks; responsive navigation, search, prompt copying and JSON handover were also checked.

The workflow review was conceptual and read-only. It did not certify a deployed multi-agent runtime, fresh full audiovisual film quality, current platform status or legal clearance. The HTML tracker manages local records; it does not authenticate approvals, inspect evidence, discover missing units or automatically infer dependencies.

Review sequence: review this HTML/PDF and the tracker; record desired changes; approve the content; then finalize the skill and create the requested private GitHub project. Public release remains a separate later decision. No repository or global skill installation was created for this edition.

Toolkit extension review: an independent agent checked all 31 department routes and 97 activity mappings. Four practical issues were corrected: budget audio routes now include picture synchronization; planning/proposal/review checks no longer instruct users to execute repairs; unverified price comparisons were removed; restore tests require the original authoring application and plugins. The 23 profiles distinguish documented features, creator experience, current cost observations and untested performance. Reelcraft is disclosed as creator-owned and currently for owner-only private use.

Project workspace review: an independent code audit found four recoverability issues. Checklist commit failures now surface in the workspace and block dependent writes and backups; unsent forms stay with their film; uploads are tracked from preparation through saving; invalid inputs can be corrected without disabling future saves. Local files remain in this browser profile. Authenticated hosted accounts and private cloud storage are a proposed next phase.

Sources and evidence boundaries

The guide adapts the supplied AI filmmaking handbook, session notes and instructor prompt sequence, along with selected Apple at the Door and Night Watch production evidence. Original documents and private working records are not bundled. Night Watch was completed and submitted; stale boxes in older tracking snapshots are not evidence that the finished film remains incomplete.

The filmmaker map separates documented practices from proposed AI adaptations. Tool capabilities and destination rules must be checked at actual production time. The Divinium rubric remains unidentified; the operational rubric is provisional.

• IndexedDB storage model
• browser storage limits
• Auth guide
• row-level security guide
• storage access controls
• Canon field-of-view reference
• Canon depth-of-field reference
• ACES encoding documentation
• Dolby authoring manual
• EBU R128
• W3C media accessibility planning
• Runway · runway.com / pricing
• LibreOffice · www.libreoffice.org
• Higgsfield · higgsfield.ai
• Higgsfield · higgsfield.ai / pricing
• Higgsfield · higgsfield.ai / how do higgsfield plans work
• VicSee · vicsee.com
• VicSee · vicsee.com / pricing
• VicSee · vicsee.com / terms of service
• Runway · runway.com
• Magnific (formerly Freepik) · www.magnific.com
• Magnific (formerly Freepik) · www.magnific.com / pricing
• ComfyUI · www.comfy.org
• ComfyUI · support.comfy.org / 9806306061 about comfy
• ComfyUI · docs.comfy.org / overview
• Reelcraft · Agentic AI Solutions · reelcraft.agenticaisolutions.tech
• FinalTake · www.finaltake.ai
• Krita · krita.org / en
• Blender · www.blender.org
• Blender · www.blender.org / features
• Blender · www.blender.org / about
• CapCut · www.capcut.com
• CapCut · www.capcut.com / how much does capcut pro cost
• DaVinci Resolve · www.blackmagicdesign.com / davinciresolve
• DaVinci Resolve · www.blackmagicdesign.com / studio
• Kdenlive · kdenlive.org
• Kdenlive · kdenlive.org / about
• Kdenlive · docs.kdenlive.org / introduction.html
• Suno · suno.com
• Suno · suno.com / pricing
• Suno · help.suno.com
• Suno · help.suno.com
• ElevenLabs · voice and SFX · elevenlabs.io
• ElevenLabs · voice and SFX · elevenlabs.io / pricing
• ElevenLabs · voice and SFX · help.elevenlabs.io / 13313564601361 Can I publish the content I generate on the platform
• ElevenLabs · voice and SFX · elevenlabs.io / eleven music model specific terms
• Audacity · www.audacityteam.org
• REAPER · www.reaper.fm
• REAPER · www.reaper.fm / purchase.php
• Freesound · freesound.org
• Freesound · freesound.org / faq
• MediaInfo · mediaarea.net / MediaInfo
• FFmpeg and ffprobe · ffmpeg.org
• FFmpeg and ffprobe · ffmpeg.org / about.html
• FFmpeg and ffprobe · ffmpeg.org / ffprobe.html
• Shutter Encoder · www.shutterencoder.com
• DGA interview, 2008
• DGA Digital Day, 2022
• ASC collaborator account
• Wētā's production account
• Dolby sound-team interview page
• DGA interview, 2006
• Team 1 — community member’s film
• Team 2
• Team 3
• Team 4
• Team 5
• Team 6
• Team 7
• Praful ji