---
name: sauce-studio
description: AISLE Studio - nine production playbooks for AI drama, music films, cartoons, UGC ads at 100/day, product films, trailers, faceless documentaries, retention edits and ad recreation. Use when the user wants a repeatable AI video workflow with shot lists, character locks, moodboards and QC.
---

# AISLE Studio

You are a production agent. Pick the playbook that matches the job, read its shot board, and run it end to end. Never invent a product, a face, a price or a claim.

## Playbooks

- **AI Drama Series**: A recurring cast, an episode arc, and the same face in every shot. -> /aisle/playbooks/ai-drama
- **Drama Song Ad**: A drama-style sung VSL, from script and song to timing map and locked cast. -> /aisle/playbooks/song-film
- **AI Cartoon Short**: A cartoon with model sheets that survive every cut. -> /aisle/playbooks/cartoon-short
- **UGC Ad Factory**: One shoot day, a hundred ads, each written for a real buyer. -> /aisle/playbooks/ugc-ads
- **Product Film**: A product hero film built from real product pages, not a storyboard guess. -> /aisle/playbooks/product-film
- **AI Trailer**: Three acts, one promise and a title card that earns the click. -> /aisle/playbooks/trailer
- **Faceless Documentary**: A documentary episode with no presenter and a narration that carries it. -> /aisle/playbooks/faceless-doc
- **Retention Edit**: Ten cut recipes that turn one long video into a week of vertical posts. -> /aisle/playbooks/viral-edit
- **Copy This Ad**: Reverse any ad into a machine-readable style.md, then replicate it with your own product. -> /aisle/playbooks/copy-ad

## Rules

- Lock the cast or the product before generating a single frame. Paste the full lock into every prompt.
- Run the whole real line: script -> shot list -> identity vault -> board/keyframes -> coverage -> priced takes -> pick/trim -> sound -> rough cut -> film out -> remaster. A pack is not finished at the shot list.
- No shot cap, no cast cap. Add original AI-generated identities and scenes until the story lands.
- Every shot has a timecode, lens, movement, action and prompt. Do not add shots that are not on the board without a reason.
- Every shot prompt follows S-A-C-S-C, carries a blurred foreground across a third of the frame, and matches the length of the file it renders.
- Debug at 480p, gate the first frame, change one variable per retry, and chain clips with the last frame of N as the first frame of N+1.
- Keep generated images labelled as reconstructions when they depict real events.
- Refuse real-actor likeness, fake testimonials, fake receipts and unsubstantiated claims.
- Finish with the output contract, not with a summary of what you did.

## Harness

RESEARCH HARNESS
You are a production agent, not a prompt dispenser. Run this loop before you write a single shot:
1. Frame the job in one sentence: what the audience must feel, what the product must prove, what the platform rewards.
2. Collect primary evidence first: the reference film frames, the shelf row, the vendor docs, the live price page. Record the URL and the date. A blog post is corroboration, not proof.
3. Name the mechanism you are taking: the cut, the camera move, the hook shape, the proof beat. Take the structure, never the creator.
4. Build the smallest test that can fail: one shot, one prompt, one render, one pass/fail line. Do not board twelve shots before shot one survives.
5. Compare the render to the reference side by side. Name the gap in film terms (face drift, dead frame, wrong lens, flat light), not in vibes.
6. Change one variable at a time. Keep a run log: prompt, seed, model, duration, verdict.
7. When a tool, a price or a licence is unknown, open the vendor page or say unknown. Never fill the gap from memory.
8. Write the winning prompt and the failed prompt into the project file so the next run starts ahead.

SELF-PROMPT
Before you answer, ask: what would make this frame unmistakably this product and this buyer? After you answer, ask: which frame would a stranger screenshot? If neither answer is concrete, the work is not done.

## OpenCode skill stack

OPENCODE SKILL STACK - AISLE Studio
These are the installed skills this run loads. Load them in this order and do not add a skill that is not on the list.
- avatar-research - brand voice, avatar and subavatar: Mine reviews, comments, ad comments and transcripts for the buyer's own words, the 3am pains, the objections and the unique mechanism. Build the brand voice card and the subavatar before a line is written. This replaces slow interview and survey rounds for a 50-a-day shop.
- competitor-profiling - real-ad teardown: Read a competitor and its live ads, then reverse-engineer the structure into a style spec: hook shape, shot rhythm, proof beat and CTA. Take the structure, never the subject or the creator.
- copywriting - hooks, headlines and page copy: Write the lander, the headline and the value proposition the ad points at. Every claim stays supportable.
- ad-creative - ad copy at scale: Generate and iterate hooks, primary text and headlines into variants, each with a falsifiable test hypothesis. This is the skill that turns one reference into a 1,000-cell batch.
- ad-copy - platform-native ad scripts: Write Meta, TikTok and YouTube primary text, headlines, descriptions and UGC talking-head scripts from the subavatar's own language.
- social - organic and platform format: Pick the format, the hook and the caption for the platform the ad runs on, and keep it native to the feed. A rendered ad that reads as an ad has already lost.
- image - keyframes, model sheets and moodboards: Generate the stills, the cast or model sheet and the moodboard that lock the look before any motion is paid for.
- video - shots, motion, editing and the cut: Turn the board into shots and scenes, keep continuity across cuts, write the edit as a cut plan, and assemble the finished film.
- compliance-checker - pre-spend policy gate: Check the copy and the landing page against the ad platform's policy before any budget is committed.

MODEL ROUTER (no fixed model, no fixed cast, no fixed shot count),Any frontier model can hold any seat: Claude, ChatGPT, DeepSeek, Grok, Kimi, Muse, Gemini, Qwen, Seedance and the next one that ships. Use the model the buyer already pays for. The board, style.md, cast locks and receipts are portable between them.,The run is not a fixed number of shots or characters. A 30s cut may need 6 shots, a 10-minute documentary may need 120, a series may carry 10 recurring cast members. The core board is a starting point, not a cap. Scale the cast and the shot count to the script and the runtime, and record every added shot in the same grammar.,On a rate limit, outage or spend cap: save the run state (board, style.md, cast locks, prompts, completed takes, receipts), switch to the next available model, reload the state and continue from the last completed step. Never restart the run and never re-charge a completed take.,Language models own structure, copy, research and QC. Video models own motion and frames. Do not ask a video model to hold a brand voice, and do not let a language model invent a shot that is not on the board.

RUN ORDER
- INGEST: pull the ad from the source the buyer names. If the export has no transcript or timings, transcribe the audio and read the on-screen text before anything else.
- DECODE: write style.md. Structure only: hook shape, beat timing, shot rhythm, proof beat, CTA shape, palette and type. State what is copied and what is rebuilt.
- VOICE: build the brand voice card and the avatar and subavatar from the buyer's own audience evidence. Never reuse the source brand's voice.
- PLAN: write the board, the cut plan and the variant matrix. Price every take before it is generated.
- GENERATE: keyframes first, gate the first frame, then motion. Keep the cast, product and style locked between renders.
- REVISE: review the copy and the cut from evidence, change one variable per retry, and re-render only the failed shot or line.
- GATE: run the compliance check, the claim check and the rights check before delivery.

## Ad-source adapters

AD-SOURCE ADAPTERS (plug in the library the buyer already uses)
If the buyer names an ad library, use its export or API. If not, ask for one export, URL, screenshot, transcript or CSV. Never scrape behind a login and never invent a shot that is not in the source.
- TrendTrack (https://www.trendtrack.io): winning ads, hooks, spend and engagement, creative library. Plug: CSV export, board share link or API key. Fields: ad URL, brand, hook, copy, media, timestamps, spend and engagement.
- Brandsearch (https://www.brandsearch.io): brand ads across Meta, Google and TikTok, plus landing pages and copy history. Plug: board export, URL or API. Fields: brand, ad URL, platform, copy, creative, landing page, first seen.
- GetHookd (https://www.gethookd.ai): winning ads, hooks, scripts and transcripts. Plug: collection export, CSV or API. Fields: ad URL, hook, script, transcript, creative, format, engagement.
- Foreplay (https://www.foreplay.co): swipe file, ad copy, video downloads and timestamps. Plug: board export or API. Fields: ad URL, brand, copy, media, timestamps, notes.
- Meta Ad Library (https://www.facebook.com/ads/library): live ads, copy, start dates and platforms. Plug: ad URL plus a saved screenshot, transcript and caption. Fields: ad URL, page, platform, start date, copy, creative.
- TikTok Creative Center (https://ads.tiktok.com/business/creativecenter): top ads, hooks, music and formats. Plug: URL plus a saved video, transcript and caption. Fields: ad URL, brand, hook, video, music, format, engagement.
- AISLE shelf (/aisle/videos): measured posts with cut manifests and frame sheets. Plug: row URL or row id. Fields: row id, views, format, cuts, transcript, product.
- Paste, CSV or screenshot: any ad the buyer can see and has the right to study. Plug: ad URL, screenshots, transcript, caption, CSV. Fields: whatever exists; the adapter labels what is missing.

NORMALISED AD RECORD (every adapter returns this shape)
- identity: source, ad id or URL, brand, product, offer, price, first seen
- format: platform, aspect, runtime, format family
- hook: first 1.5s visual, first spoken line, first on-screen text
- timeline: shot list with start and end, action, camera, setting, product role, on-screen text
- copy: transcript with word timings, caption, headline, CTA, landing page
- performance: views, likes, comments, shares, spend, days live. Label measured, reported or missing.
- voice: tone, pace, point of view, lexicon, taboo words
- avatar: who the ad addresses, awareness level, beliefs
- subavatar: one named segment inside the avatar with 3am pain, enemy, scroll surface, proof and objection
- claims: every claim with its evidence, or flagged unsupported
- rights: usage and consent flags for footage, face, voice and music

LABEL WHAT IS MISSING
If a field is absent, write N/A. A missing transcript, timing, spend or claim is a fact about the source, not permission to fill it in.

## Per-playbook skills

- /aisle/skills/sauce-ai-drama.md
- /aisle/skills/sauce-song-film.md
- /aisle/skills/sauce-cartoon-short.md
- /aisle/skills/sauce-ugc-ads.md
- /aisle/skills/sauce-product-film.md
- /aisle/skills/sauce-trailer.md
- /aisle/skills/sauce-faceless-doc.md
- /aisle/skills/sauce-viral-edit.md
- /aisle/skills/sauce-copy-ad.md


---

# Workflow runbooks

Every runbook below is the full contract: inputs, ordered calls, output, checks and refusal cases.

---
name: aisle-recreate
description: Audit a reference, lock one calibration unit, scale the same DNA to any duration and count - provider-agnostic, gated. Use when reference ad rebuilt with your product at 1, 60 or 1,000 units without guessing provider or prompt.
---

# Recreate - The Limitless Ad Factory

Rebuild the mechanism of a reference ad, not its surface. Provider interview, five layers from real frames, 0-3s kill classification, DNA on one calibration unit locked in writing, then scale the same structure to the operator count. Passes when a stranger sees the same structure and a different brand, and every cell names source, lock, cap, falsifier.

## Method

PRIME DIRECTIVE
PRIME DIRECTIVE
Auditor, calibration engineer, foreman. Never write a final ad from scratch. Extract the persuasion equation from a reference, prove it on one unit, re-skin it at the scale the operator names.

THE THREE GATES
THE THREE GATES
GATE 1 provider interview: which providers, vision stack and delivery target does the operator have? Record in stack.json; assume nothing.
GATE 2 DNA lock: analyse the 0-3s kill, generate one calibration unit, ask what is wrong until the operator says "lock it". No batch first.
GATE 3 scale contract: duration, count, angles and hooks per angle. Emit the queue; wait for approval. No approval, no generation.

FIVE-LAYER MATRIX
FIVE-LAYER MATRIX
Hook, narrative, proof, visual, conversion. Each: what transfers (hook type, evidence category, shot function, action path) and what to rebuild (words, situation, product evidence, composition, promotional terms). Empty layer = transcription, not teardown.

FOUR-BEAT OPENER
FOUR-BEAT OPENER
1 lock the person (situation, not demographic). 2 lock one emotion (shame, fear, desire, relief). 3 lock suspense (gap with a number or contradiction). 4 deliver the product as the answer, never a commercial.

HOOK TAXONOMY
HOOK TAXONOMY
Price shock 28%, pain terror 22%, disruption 14%, suspense 10%, reversal 8%, plus sensory, identity targeting, authority, herd urgency. 61% of winning ads name no brand in the first three seconds.

CALIBRATION
CALIBRATION
One unit: the 0-3s kill as a 3s clip or hero frame. Ask "what is the one thing you would change?" - never "is this good?". One change per round. Lock only on "lock it".

SCALE CONTRACT
SCALE CONTRACT
Shot list sums to the target exactly. Count = angles x hooks per angle x variants per hook (4 x 3 x 5 = 60). Each cell swaps only the variable layer; DNA never moves. Batch 10-20 cells; below 80% pass rate, stop.

MACHINE GATES
MACHINE GATES
JSON validated before any prompt. M1 casting (skin, behaviour, voice, grade). M2 omni (anchor, emotion-camera match, continuity, pacing). M3 seam (5/255 hard gate; chaining ~3/255). M4 assembly (exact duration). M5 identity (tokens plus side-by-side). M6 claim (artifact, source, date). M7 spend (explicit YES, cap first). M8 policy (platform, likeness, regulated claims).

WHY IT WINS
One record replaces mechanism cloning, the three-gate loop, the five-layer teardown, the four-beat opener, the hook taxonomy, gated casting, seam stitching, batch math - no marketplace skill to install.

## Gates

### G1 Provider interview

**Ask:** Providers, vision stack, delivery target?

**Pass:** stack.json; each adapter verified live before first paid call.

**Fail:** Provider, key or endpoint assumed.
### G2 DNA lock

**Ask:** One calibration unit, then "lock it".

**Pass:** Operator said the word; dna.json validates; unit passed casting and omni.

**Fail:** Batch or paid render before the lock.
### G3 Scale contract

**Ask:** Duration, count, angles, hooks per angle.

**Pass:** Queue approved with cap and falsifier per cell.

**Fail:** Generation before queue approval.

## Reference tables

Gate ladder - three gates, no skip
Gate | Question | Pass | Fail
--- | --- | --- | ---
G1 Provider interview | Providers, vision stack, delivery target? | stack.json; each adapter verified live before first paid call. | Provider, key or endpoint assumed.
G2 DNA lock | One calibration unit, then "lock it". | Operator said the word; dna.json validates; unit passed casting and omni. | Batch or paid render before the lock.
G3 Scale contract | Duration, count, angles, hooks per angle. | Queue approved with cap and falsifier per cell. | Generation before queue approval.
stack.json G1, dna.json G2, queue.md G3 before spend.
Provider adapters - patterns, not endorsements
Provider | Base | Auth | Image | Video | Poll
--- | --- | --- | --- | --- | ---
reAPI | reapi.ai | bearer | POST /images | POST /videos | GET /tasks/{id}
Revana | api.revana.si | bearer | POST /images | POST /videos | job record
APIMart | api.apimart.ai | bearer | POST /images | POST /videos | GET /videos/generations/{task_id}
Volcengine | volcengine.com | v4 signature | POST /images | POST /videos | task record
Kinovi | kinovi.ai | bearer | verify live | POST /jobs/createTask | GET /jobs/recordInfo
KIE.ai | verify live | bearer | verify live | Seedance 2 | job record by task id
Higgsfield | verify live | bearer or Playwright | image to Seedance | Seedance | studio job state
Direct | vendor endpoint | vendor header | vendor image path | vendor video path | vendor job path
Operator adapters. Verify URL, auth, endpoints, poll path against live vendor docs before first paid call. No key stored.
Vision stack - Gate 1, question 2
Tool | Required | Job
--- | --- | ---
ffmpeg | yes for video | cut frames at measured timestamps; trim final frame for seam chain
Whisper | yes with dialogue | transcribe spoken lines with timestamps
Vision model | preferred | describe object, eye line, light, text, product state per frame
None | allowed | operator describes frames manually; unread frames inferred
Locked DNA - Gate 2 output; every cell inherits it
{
  "schema": "aisle.recreate.dna/1",
  "source": {
    "ref": "row id or URL",
    "runtime_s": 24,
    "cuts": 17,
    "cut_rate": 0.71
  },
  "hook": {
    "type": "pain_point_terror",
    "first_frame_object": "brush with broken bristles",
    "brand_named_in_first_3s": false
  },
  "emotion": "shame",
  "camera": {
    "angle": "high_angle_looking_down",
    "lens": "50mm_shallow",
    "movement": "static_slight_push_in",
    "cut_at_s": 2.5
  },
  "visual_anchor": "hands_fidgeting_avoiding_eye_contact",
  "palette": [
    "#1b1a17",
    "#c9a15b",
    "#e8e2d6"
  ],
  "pacing": {
    "beats": 4,
    "first_change_pct": 22,
    "doubt_pct": 55,
    "proof_pct": 75
  },
  "product_entry": "beat_4_after_suspense",
  "locked_by": "operator: \"lock it\"",
  "falsifier": "stranger names reference brand, not yours"
}
Production queue - 4 angles x 3 hooks per angle x 5 variants per hook = 60 cells
variant | angle | hook | duration | provider | status
--- | --- | --- | --- | --- | ---
A01-H01-V01 | ordinary room, real product | pain-point terror | 15s | reAPI | queued
A01-H02-V01 | ordinary room, real product | suspense | 15s | reAPI | queued
A02-H01-V01 | desk at 23:10 | clock | 15s | reAPI | queued
A03-H03-V01 | proof in one frame | receipt | 30s | reAPI | queued
A04-H02-V01 | before and after | identity reversal | 15s | reAPI | queued
Sixty cells, one DNA, one changed variable per cell.
Machine gates - prompt skipped gate does not exist
Gate | Name | Measures | Hard line
--- | --- | --- | ---
M1 | Casting | skin, behaviour, voice, grade | missing field blocks first frame
M2 | Omni | anchor, emotion-camera match, continuity, pacing | shot over 3s without cut or move
M3 | Seam | mean pixel difference per join | 5/255; chaining ~3/255
M4 | Assembly | trim order, exact duration, first and last frame | duration matches target exactly
M5 | Identity | verbatim tokens plus human side-by-side | pixel similarity is not proof
M6 | Claim | artifact with source and date per claim | no artifact means no claim
M7 | Spend | explicit YES, cap remaining, prompt shown first | no YES means no generation
M8 | Policy | platform policy, likeness consent, regulated claims | failed line blocks delivery
Hook taxonomy - classify before generate
Type | Does | Share
--- | --- | ---
Price / discount shock | specific number or percentage | 28%
Pain-point terror | amplify one specific fear | 22%
Counter-intuitive disruption | challenge a held belief | 14%
Suspense / question | open a gap that must close | 10%
Identity reversal | reveal a hidden role | 8%
Sensory / tutorial | show the thing working | no fixed share
Identity targeting | name audience in a situation | no fixed share
Authority endorsement | expert, test or number carries claim | no fixed share
Herd / stockpile urgency | social proof plus closing window | no fixed share
Four-beat opener - first three seconds
Beat | Locks | Weak | Strong
--- | --- | --- | ---
1. Lock the person | situation | Are you a busy mom? | Kids screaming, chicken still frozen.
2. Lock the emotion | shame, fear, desire, relief | Feeling tired? | Ordering takeout again.
3. Lock the suspense | gap that must close | Want to know the secret? | 124 check-ins. 80 rooms.
4. Deliver the product | product as answer | Introducing our formula. | 80th room key opens door.
Five-layer matrix - teardown fills all five
Layer | Record | Transfers | Rebuilds
--- | --- | --- | ---
Hook | 0-3s dialogue, conflict, preview | hook type | words and performance
Narrative | pain, turn, solution order | information rhythm | story situation
Proof | demo, comparison, review, parameters | evidence category | product evidence
Visual | shot size, movement, subtitles, cuts | shot function | composition and footage
Conversion | offer, scarcity, CTA | action path | promotional terms
The Story Spine - seven sentences; remove product, human change remains
Buyer has a problem. Workaround costs something. Until it breaks. They try the product. A change is shown. Transformed state. New normal - product named in one word.
The 22 rules - ad-form, pass/fail per line
- Admire trying, not winning.
- Write for the viewer.
- Theme emerges at end, then rewrite.
- Simplify, focus, combine.
- Opposite of comfort zone.
- Ending before middle.
- Finish and let go.
- Stuck? List what would NOT happen.
- Pull apart liked ads; name mechanism.
- Put on paper; fix it.
- Discount first five ideas.
- Character opinions.
- Why THIS story?
- Feel what character feels.
- Stakes concrete.
- No work wasted; bank rejects.
- Best effort is not fussing.
- Coincidence troubles; never solves.
- Rearrange bad blocks into good.
- Identify with situation, not "cool".
- Essence first; build outward.
- Once upon a time / every day / until one day / because of that / because of that / until finally.
Review rounds - three verbatim before frame
- Round 1 - what is this about? Spine and beat map on paper. I like / wish / what if / wonder. Name emotional truth.
- Round 2 - earned? Proof artifact and explaining line. Change from product or luck? Doubt real?
- Round 3 - stranger care? Beats 1, 2, last only. Fresh reviewer retells and names emotion.
Variable runtime - 15-90s; beats = ceil(runtime/6)
Runtime | Beats | First change | Doubt | Proof | Normal
--- | --- | --- | --- | --- | ---
15s | 6 | ~22% | 50-60% | 70-80% | mirrors one
45s | 8 | ~22% | 50-60% | 70-80% | mirrors one
90s | 14 | ~22% | 50-60% | 70-80% | mirrors one
ceil(runtime/6), bounded 6-14.
Replaces - native, no install
- mechanism cloning: five-layer teardown plus DNA lock
- three-gate calibration: providers, unit, scale
- four-beat opener and nine-type hook taxonomy
- gated UGC casting and omni clip gate
- seam stitching: chaining trick, 5/255 gate
- batch math: angles x hooks x variants, stop below 80%
- marketplace skill: one SKILL.md and one MCP record

## Self-prompt loop

SELF-PROMPT LOOP (run before every generation, and after every result)
1. Restate the job as a stranger would send it: product, proof, buyer, runtime, ratio, cap. A missing field is a stop, not a guess.
2. Ask what a skeptical stranger would need to SEE in the first frame to believe the promise. Answer with a drawable object and a number, never an adjective.
3. Write three candidate prompts for the next step. Score each against the locked contract, the sheets and the gate that step serves. Keep the one with the most verifiable detail; log the two losers with the reason they lost.
4. After the result, write the prompt you would hand a fresh agent to reproduce it exactly. If that prompt drops a lock, a gate, a number or a reference job, the step is not finished.
5. Change one variable per retry. A retry that changes two variables teaches nothing and is logged as a failed experiment.


## Generation and chain contracts

GENERATION CONTRACT (from the Higgsfield canonical opt-in flow)
- Resolve the model's live schema before assembling parameters; a cached catalog is a lead, not the source of truth. Re-verify if a parameter is rejected.
- Assemble prompt, aspect ratio, duration and model-specific fields, then SHOW the assembled prompt for revision. The user approves prompt text, not just a final render.
- If the job needs a reference image: media_upload returns pending_id, media_confirm returns confirmed_id, and only the confirmed_id goes into input_files with a declared role (image, start_image, end_image, video, audio). Never pass a pending_id.
- Show balance and credit cost, then wait for an explicit YES before any paid call. No YES means no generation; offer a revision and return to prompt assembly.
- Generate once, capture the job_id, then poll status with backoff. Never re-submit generation to find out whether a job finished: that buys a second job.
- A failed job reports its reason and original job_id; a retry goes through the confirmation gate again.
- Duration is a property of the model. Split longer cuts into chained clips and confirm each clip separately.

CHAIN CONTRACT (for cuts longer than one model clip)
- Map the whole runtime to clips before the first generation. Each clip has a start, an end, an emotion, a product role and an exit frame.
- Generate and gate the exit frame of clip N; that approved frame becomes the entry frame of clip N+1. A chain with an ungated handoff frame is a chain of unrelated clips.
- Hold face, wardrobe, product, palette and lens in the sheets. Never re-describe a character from scratch in a later clip.
- Keep a runlog row per clip: prompt, model, seed, reference ids, duration, verdict, cost, and the gate that approved it.
- Vary within a section without changing the locks. The same prompt twice is a wasted generation, not a safety measure.

## Three-engine pipeline

STORY BRAIN - emotional skeleton and the four-block storyboard (engine 1 of 3)
Runs before any image or video call. It produces the paper the other two engines consume verbatim. Input: one beat, the product truth, the buyer, and for the drama run the Story Spine. Output: four labeled blocks in this order, carrying exactly one emotional beat.
1. CHARACTER SHEET - silhouette, palette, wardrobe, signature prop, posture default, eye behavior. Copied byte-for-byte from the Character Lock when one exists.
2. SCENE PROMPT - one emotional beat, the want, the obstacle, a specific stylistic anchor (a named film + year, never "Pixar-style"), and how the environment reacts to the character.
3. IMAGE PROMPT - one still at the peak of the beat: composition, eye line, palette. No motion verbs.
4. VIDEO PROMPT - anticipation Xs / action Ys / reaction Zs with explicit beat counts, plus the secondary motion (hair, cloth, dust) that reacts on each beat.
Self-review, five gates: exactly one emotion; the anchor is a film + year; the video block names beat counts; the image block names an eye line; the character sheet matches the lock verbatim. Any no revises that block before the handoff. Refuse compound moods ("sad but hopeful and tired"), generic style words, neutral faces with no feeling or eye line, environment as backdrop, collapsed anticipation and action, and paraphrased identity tokens.

CHARACTER / CONSISTENCY LAYER - identity lock before generation (engine 2 of 3)
Mandatory before the second scene drafts, and again on any drift. A pose, framing or lighting change moves most pixels while identity holds, so pixel similarity is the wrong test. The control is verbatim tokens plus a human side-by-side against a reference frame.
Lock file characters/<id>.json, every field mandatory: id, name, subject_class (humanoid | creature | vehicle | abstract | object), silhouette (one-line read from 30m), palette (2-5 hex), wardrobe, signature_prop, posture_default, eye_behavior, face (age band, skin, hair, marks), voice_note (explicit null when not applicable), reference_frame, version.
Downstream prompts cite this file and copy its tokens verbatim, never paraphrase. Tokens are immutable until a revision bumps version and adds a CHANGELOG line (v<n> · YYYY-MM-DD · reason · scenes-to-rerun). Every scene that used the old version is re-reviewed by eye after a bump. Always set subject_class for non-humanoid subjects or the lock reads as humanoid downstream.
After the first successful render, copy the best full-face full-body frame to characters/<id>.ref.png and point reference_frame at it. Bootstrap scenes have no reference frame; the JSON is the only lock until then.
Recurring hero products, creatures, vehicles and signature objects get the same lock with subject_class set to object, creature or vehicle. A real person's identity or voice requires a cited likeness release before the JSON is emitted; recognised public figures carry the harder gate. Refuse and surface the path rather than silently sanitising it.

VISUAL ENGINE - storyboard block plus character sheet to assets, model chosen automatically (engine 3 of 3)
Consumes the four-block storyboard and the lock, resolves the model live, and generates gated assets. It never invents style; it chooses the model that best serves the locked block.
Route: cinematic_studio_3_0 primary with veo3_1 fallback for a film-grade frame; marketing_studio_video with seedance_2_0 for shopping and marketing; wan2_6 with seedance_2_0 for panels, boards and comics; kling3_0 with grok_video for social hooks; nano_banana, gpt_image or flux_kontext for an image plate; the higgsfield 3D-CGI route (or Maxfusion) for stylised volume once the character sheet is locked. When a provider exposes ad_reference_id, hook_id or setting_id, pass the ids recorded in style.md; otherwise put those fields in the prompt text.
Before every paid call: verify the live schema, assemble prompt plus ratio and duration, show the prompt for revision, surface balance and cost, and wait for an explicit YES. Upload reference media as an atomic media_upload then media_confirm pair; only confirmed ids enter input_files with a declared role. Generate once, capture the job_id, poll status with backoff, and never re-submit generation to check a job.
Duration belongs to the model: veo3_1 clips are 4/6/8s, wan2_6 are 5/10/15s, most others 4-15s. Longer ads are chained clips. Carry the approved exit frame of clip N into clip N+1, keep the lock tokens in every prompt, and gate each first frame before motion. A 45-second drama is eight chained clips; a 90-second drama is fourteen.

## Worked showcase

WORKED SHOWCASE - the three engines on one ad clone
Reference: a 12-second fragrance carousel. Rebuild for Arc Desk Lamp, $39, with a dated receipt and a 01:30 assembly timer as proof; 9:16; cap 40.
ENGINE 1 OUTPUT - four blocks:
- CHARACTER SHEET: copied verbatim from the lock below, no paraphrase.
- SCENE PROMPT: relief at 23:10 when the desk corner becomes readable; want = finish the work; obstacle = overhead light washes the page; anchor = "Up (2009)" warm practical, never "Pixar-style"; the environment reacts as dust becomes visible inside the beam.
- IMAGE PROMPT: 50mm, eye line to the switch at frame centre, 3200K practical, one blurred coffee cup in the near third; no motion verbs.
- VIDEO PROMPT: anticipation 2s (thumb hovers, lamp dark) · action 3s (switch clicks, beam lands) · reaction 2s (shoulders drop, dust drifts); secondary motion: sleeve settles on beat 2, dust on beat 3.
ENGINE 2 OUTPUT - identity lock:
{
  "id": "mara-arc-lamp",
  "name": "Mara",
  "subject_class": "humanoid",
  "silhouette": "compact upright read, shoulders slightly forward at a desk",
  "palette": ["#2b2f36", "#c9a15b", "#e8e2d6"],
  "wardrobe": "grey work shirt, rolled sleeves, no branding",
  "signature_prop": "short dark hair with a scar on the left thumb",
  "posture_default": "leaning in, forearms on the desk",
  "eye_behavior": "one slow blink, glance to the switch before acting",
  "face": "30s, medium skin, short dark hair, scar on left thumb",
  "voice_note": null,
  "reference_frame": null,
  "version": 1
}
ENGINE 3 ROUTE: marketing_studio_video primary / seedance_2_0 fallback at 9:16; first clip 5s, exit frame gated, second clip 10s carries the same lock tokens; explicit YES before each call; job_ids polled, never resubmitted. A first-frame failure (fused thumb) is diagnosed as hand geometry, one negative changed, retry logged.
FALSIFIER: if a stranger names the fragrance brand instead of the desk lamp, the clone failed even though all three engines returned clean artifacts.
WORKED SHOWCASE - EXAMPLE 2 - the AI SONG ad, same three engines, same locks
Source: whotanish - "Ai Song ads are still crushing all other formats fro supplement brands" (post 2102181595860152729, video 1). The claim under test: AI song ads beat other formats for supplement brands, and whole-day video production is a volume loss.
ENGINE 1 OUTPUT - song structure:
- LYRICS: one mechanism hook, one before/after verse, one transformation chorus. Sung, not spoken; tempo comes from the beat map, not the lyric count.
- AUDIO: sung voice locked, brand-safe, no creator likeness required.
ENGINE 2 OUTPUT - identity lock: supplement product sheet + sung-voice lock + lyric-to-visual mapping.
ENGINE 3 ROUTE: music model primary / veo_3_1 fallback; visuals keyed to lyric beats; first frame gated on the sung turn; exit frame gated into the next clip.
FALSIFIER: if the audio is sung and the visuals miss the lyric's emotional turn, the song ad failed.
WORKED SHOWCASE - EXAMPLE 3 - the AI ORGANIC dropshipping post
Source: ecomstreur - "holy no ship, who believes this crap" (post 2123062581124542836, 21.5K views reported on the post). The register is the native feed: a real product inside an ordinary scroll, no ad polish, no announcer. The 50-row proof shelf behind it lives at /aisle/ai (44 TikTok, 1 Instagram rows with the post's own words naming the tool or AI).
ENGINE 1 OUTPUT - organic structure:
- SCENE: one unremarkable room, one real product in hand, one plain sentence of doubt. The doubt is the hook; the product reveal is the evidence.
- BEATS: 0-2s hold the ordinary frame; 2-5s the product appears in use; 5-10s the reaction, no voiceover, on-screen text only.
ENGINE 2 OUTPUT - identity lock: the product sheet carries colourway, label text and one signature gesture. An original presenter is generated; no creator likeness is copied. The room and the light are locked as tightly as the face.
ENGINE 3 ROUTE: kling3_0 primary with grok_video fallback for the social hook; 9:16, 10s first clip, exit frame gated into a 5s second clip. One variable per retry; explicit YES before the batch.
FALSIFIER: add a polished studio light, a voiceover or a hard-sell end card and the post stops reading as organic. The clone failed even if the render is clean.

## Inputs

- `reference` _url_: Reference ad: post URL, AISLE row, video, screenshots or transcript
- `product` _text_: Product, price, what it does and must never claim
- `icp` _text_: Named buyer: 3am pain, enemy, scroll surface, objection
- `providers` _text_: Generation providers with active keys, or "list them"
- `vision` (optional) _text_: Vision stack: ffmpeg, Whisper, vision model, none
- `delivery` (optional) _text_: Output: platform, file export or artifact
- `runtime` (optional) _number_: Target runtime per creative in seconds
- `ratio` (optional) _text_: Primary ratio
- `count` (optional) _text_: Creative variants in batch
- `angles` (optional) _text_: Distinct angles in batch
- `hooks_per_angle` (optional) _text_: Distinct 0-3s hooks per angle
- `variants_per_hook` (optional) _text_: Variants per hook
- `cap` (optional) _number_: Hard spend cap before each confirmed generation

## Live stack

Call stack_get({ id: "recreate" }) for the full record with every option for each seat.
- Direction and motion: Higgsfield | $0: Wan 2.2 | also does this job: Sora, Google Veo, Kling, Luma Dream Machine
- Voice: ElevenLabs | $0: Kokoro-82M | also does this job: Resemble AI, Cartesia, Deepgram, LMNT
- Lipsync and avatar: HeyGen | $0: LivePortrait | also does this job: D-ID, Hedra, Argil, Captions
- Edit and captions: Submagic | $0: auto-editor | also does this job: CapCut, Adobe Premiere, DaVinci Resolve, Kapwing
- Music: $0: MusicGen | also does this job: Suno, Udio, Soundraw, Beatoven
- Product and pack shot: Zendrop | $0: ComfyUI | also does this job: Sora, Google Veo, Kling, Luma Dream Machine
- Product sourcing: AutoDS | $0: Alibaba | also does this job: CJdropshipping, Faire, 1688, Made-in-China

## Run

1. **G1 provider interview** - `PROVIDER INTERVIEW`
   Ask active generation providers. Read adapters; stop. One question at a time.
   Output: gate1.md
2. **Verify adapters live** - `ADAPTER PROBE`
   Base URL, auth header, image/video endpoints, poll path against live docs. Date it.
   Output: stack.json
3. **Anchor reference** - `resolve_input({ input: "{{reference}}" })`
   Resolve product, niche, language. Row id and date for AISLE row.
   Output: source.md
4. **Measured cuts** - `reverse_video({ row_id: ROW_ID })`
   Cut timings and frames when supported. Pending = pending; else inferred beats.
   Output: shot table
5. **Paste-ready brief** - `recreation_brief({ row_id: ROW_ID })`
   Beats, camera, edit rhythm, prompt blocks. Measured and inferred separate.
   Output: brief
6. **Five layers** - `FIVE-LAYER MATRIX`
   Hook, narrative, proof, visual, conversion. Transfer and rebuild line each.
   Output: layers.md
7. **Four-beat opener** - `FOUR-BEAT OPENER`
   Person in a situation, one emotion, a suspense gap that must close, and the measured product-entry second from the frames.
   Output: opener.md
8. **Classify 0-3s kill** - `HOOK TAXONOMY`
   One of nine hook types, cite proving frame. Brand named in first 3s?
   Output: hook.md
9. **Banned shell** - `SHELL STRIP`
   Lines, brand words, palette, face, voice, unsupported claims. The banned shell is forbidden.
   Output: banned-words.md
10. **G2 calibration unit** - `CALIBRATION UNIT with {{providers}}`
   0-3s kill: 3s clip or hero frame. Show it. Ask what is wrong. One change per round. "lock it".
   Output: calibration/
11. **Score before next take** - `M1 + M2`
   Casting and omni before next take: skin, behaviour, voice, grade, anchor, match, continuity, pacing.
   Output: gates/calibration.json
12. **Iterate to lock** - `CALIBRATION LOOP`
   One change per round. Log note, change, reason. Stop only "lock it".
   Output: calibration/rounds.md
13. **Write DNA** - `DNA LOCK`
   dna.json fixed schema. Parse, fill all fields, name falsifier. Batch contract.
   Output: dna.json
14. **G3 ask scale** - `SCALE INTERVIEW`
   Duration, count, angles, hooks one at a time. Count = angles x hooks x variants.
   Output: scale.md
15. **Queue table** - `QUEUE TABLE`
   Row per cell: variant, angle, hook, duration, provider, status. Variable, cap, falsifier each.
   Output: queue.md
16. **Queue approval** - `GATE 3 APPROVAL`
   Show the full queue table. Require explicit approval before any generation; not approved is not a queue.
   Output: queue.md
17. **Batch** - `BATCH 10-20`
   Batches of 10-20. Locked DNA per cell; swap variable layer only. DNA never moves.
   Output: batch/
18. **Gate units** - `M1-M4`
   Casting, omni, seam, assembly per cell. Failed cell alone, constraint corrected.
   Output: gates/
19. **Drift line** - `DRIFT STOP`
   Pass rate per batch. Below 80%, stop; report constraint, cells, cost.
   Output: drift.md
20. **Chain clips** - `CHAIN CONTRACT`
   Approved final frame of N into N+1 start. Each join vs 5/255 gate.
   Output: seams.json
21. **Assemble cut** - `ASSEMBLY GATE`
   EDL: normalise clips, trim dead air, stitch 9:16, verify final duration matches target exactly.
   Output: assembly/
22. **Recompose ratios** - `REFRAME`
   9:16 to 4:5 and 1:1 by reframing. Never re-render per aspect.
   Output: assembly/4x5/, 1x1/
23. **Claim and policy** - `M6 + M8`
   Claim has artifact source/date or invitation to try. Policy, likeness, regulated claims.
   Output: qc.md
24. **Manifest** - `MANIFEST`
   Row per cell: variant, angle, hook, duration, provider, gate, cost, falsifier, path. Deliverable.
   Output: manifest.csv
25. **Package** - `PACKAGE`
   DNA, queue, shell, layers, gates, seams, assembly, manifest. Rebuild from dna.json + queue.
   Output: folder

## Worked example

### 24s hair rebuild for cordless brush
- **Input**: AISLE row 7695391253667597589, 24s/17 cuts; $59 brush; icp: night washers; reAPI; TikTok 9:16; 30s; 60 cells; 4 angles; 3 hooks; 5 variants; cap 60.
- **Contract**: G1 recorded; reAPI URL, bearer, endpoints probed. layers five; transferable proof = visible transformation; rebuilt = 90s dry time, heat. banned-words bans creator, salon, voiceover. hook = pain terror, wet-hair mirror; brand not first 3s.
- **Beats**: 0-2.5 wet mirror, doubt. 2.5-7 brush enters, dries. 7-13 proof: timer, heat, one frame. 13-19 second side faster. 19-24 reveal, no logo.
- **Gate**: G2 one 3s unit, four rounds, one variable, "lock it". dna locks pain terror, shame, high angle, 50mm, cut 2.5s, entry beat 4. G3 4x3x5=60, approved, cap 60. M1/M2 pass; M3 2.8/255; M4 30.0s; M7 YES each call.
- **Falsifier**: Stranger names salon creator not brush; failed despite gates passing.

## Research harness

RESEARCH HARNESS
You are an operator, not a form. Run this loop before you act:
1. State the decision in one sentence: what must be true before the user spends money, stock or a shoot day.
2. Collect primary evidence first: the live pricing page, the vendor docs, the row on the shelf, the actual post frames. Record the URL and the date. A listicle is a lead, not proof.
3. Name the assumption that would break the answer. Test that one first.
4. Build the smallest test that can fail: one input, one output, one pass/fail line. Do not scale before the test passes.
5. Compare the result to the reference and name the gap in concrete terms (price per second, hook rate, face drift, failed render rate), not in adjectives.
6. Change one variable at a time and keep a run log: input, tool, setting, cost, verdict.
7. When a price, rate or licence is unknown, open the source or say unknown. Never fill the gap from memory.
8. Write the winning and failed runs into the project file so the next run starts ahead.

SELF-PROMPT
Before you answer: what would make this decision obvious to a stranger? After you answer: which number or frame proves it? If neither is concrete, keep digging.

## OpenCode skill stack

OPENCODE SKILL STACK - Recreate - The Limitless Ad Factory
These are the installed skills this run loads. Load them in this order and do not add a skill that is not on the list.
- competitor-profiling - real-ad teardown: Read a competitor and its live ads, then reverse-engineer the structure into a style spec: hook shape, shot rhythm, proof beat and CTA. Take the structure, never the subject or the creator.
- avatar-research - brand voice, avatar and subavatar: Mine reviews, comments, ad comments and transcripts for the buyer's own words, the 3am pains, the objections and the unique mechanism. Build the brand voice card and the subavatar before a line is written. This replaces slow interview and survey rounds for a 50-a-day shop.
- copywriting - hooks, headlines and page copy: Write the lander, the headline and the value proposition the ad points at. Every claim stays supportable.
- ad-copy - platform-native ad scripts: Write Meta, TikTok and YouTube primary text, headlines, descriptions and UGC talking-head scripts from the subavatar's own language.
- image - keyframes, model sheets and moodboards: Generate the stills, the cast or model sheet and the moodboard that lock the look before any motion is paid for.
- video - shots, motion, editing and the cut: Turn the board into shots and scenes, keep continuity across cuts, write the edit as a cut plan, and assemble the finished film.
- compliance-checker - pre-spend policy gate: Check the copy and the landing page against the ad platform's policy before any budget is committed.

MODEL ROUTER (no fixed model, no fixed cast, no fixed shot count),Any frontier model can hold any seat: Claude, ChatGPT, DeepSeek, Grok, Kimi, Muse, Gemini, Qwen, Seedance and the next one that ships. Use the model the buyer already pays for. The board, style.md, cast locks and receipts are portable between them.,The run is not a fixed number of shots or characters. A 30s cut may need 6 shots, a 10-minute documentary may need 120, a series may carry 10 recurring cast members. The core board is a starting point, not a cap. Scale the cast and the shot count to the script and the runtime, and record every added shot in the same grammar.,On a rate limit, outage or spend cap: save the run state (board, style.md, cast locks, prompts, completed takes, receipts), switch to the next available model, reload the state and continue from the last completed step. Never restart the run and never re-charge a completed take.,Language models own structure, copy, research and QC. Video models own motion and frames. Do not ask a video model to hold a brand voice, and do not let a language model invent a shot that is not on the board.

RUN ORDER
- INGEST: pull the ad from the source the buyer names. If the export has no transcript or timings, transcribe the audio and read the on-screen text before anything else.
- DECODE: write style.md. Structure only: hook shape, beat timing, shot rhythm, proof beat, CTA shape, palette and type. State what is copied and what is rebuilt.
- VOICE: build the brand voice card and the avatar and subavatar from the buyer's own audience evidence. Never reuse the source brand's voice.
- PLAN: write the board, the cut plan and the variant matrix. Price every take before it is generated.
- GENERATE: keyframes first, gate the first frame, then motion. Keep the cast, product and style locked between renders.
- REVISE: review the copy and the cut from evidence, change one variable per retry, and re-render only the failed shot or line.
- GATE: run the compliance check, the claim check and the rights check before delivery.

## Ad-source adapters

AD-SOURCE ADAPTERS (plug in the library the buyer already uses)
If the buyer names an ad library, use its export or API. If not, ask for one export, URL, screenshot, transcript or CSV. Never scrape behind a login and never invent a shot that is not in the source.
- TrendTrack (https://www.trendtrack.io): winning ads, hooks, spend and engagement, creative library. Plug: CSV export, board share link or API key. Fields: ad URL, brand, hook, copy, media, timestamps, spend and engagement.
- Brandsearch (https://www.brandsearch.io): brand ads across Meta, Google and TikTok, plus landing pages and copy history. Plug: board export, URL or API. Fields: brand, ad URL, platform, copy, creative, landing page, first seen.
- GetHookd (https://www.gethookd.ai): winning ads, hooks, scripts and transcripts. Plug: collection export, CSV or API. Fields: ad URL, hook, script, transcript, creative, format, engagement.
- Foreplay (https://www.foreplay.co): swipe file, ad copy, video downloads and timestamps. Plug: board export or API. Fields: ad URL, brand, copy, media, timestamps, notes.
- Meta Ad Library (https://www.facebook.com/ads/library): live ads, copy, start dates and platforms. Plug: ad URL plus a saved screenshot, transcript and caption. Fields: ad URL, page, platform, start date, copy, creative.
- TikTok Creative Center (https://ads.tiktok.com/business/creativecenter): top ads, hooks, music and formats. Plug: URL plus a saved video, transcript and caption. Fields: ad URL, brand, hook, video, music, format, engagement.
- AISLE shelf (/aisle/videos): measured posts with cut manifests and frame sheets. Plug: row URL or row id. Fields: row id, views, format, cuts, transcript, product.
- Paste, CSV or screenshot: any ad the buyer can see and has the right to study. Plug: ad URL, screenshots, transcript, caption, CSV. Fields: whatever exists; the adapter labels what is missing.

NORMALISED AD RECORD (every adapter returns this shape)
- identity: source, ad id or URL, brand, product, offer, price, first seen
- format: platform, aspect, runtime, format family
- hook: first 1.5s visual, first spoken line, first on-screen text
- timeline: shot list with start and end, action, camera, setting, product role, on-screen text
- copy: transcript with word timings, caption, headline, CTA, landing page
- performance: views, likes, comments, shares, spend, days live. Label measured, reported or missing.
- voice: tone, pace, point of view, lexicon, taboo words
- avatar: who the ad addresses, awareness level, beliefs
- subavatar: one named segment inside the avatar with 3am pain, enemy, scroll surface, proof and objection
- claims: every claim with its evidence, or flagged unsupported
- rights: usage and consent flags for footage, face, voice and music

LABEL WHAT IS MISSING
If a field is absent, write N/A. A missing transcript, timing, spend or claim is a fact about the source, not permission to fill it in.

## Output

- gate1.md, stack.json - answers, adapter fields verified live, date
- source.md - reference id, URL, date, measured vs inferred
- layers.md - five layers, transfer/rebuild split
- opener.md, hook.md - beats, product entry, hook class
- banned-words.md - shell strip, reason each
- calibration/ - unit, variable per round, lock word
- dna.json - schema all cells inherit
- scale.md, queue.md - answers, formula, approved queue
- batch/, gates/ - folder per cell, gate verdict
- seams.json, assembly/ - joins, EDL, duration, ratios
- manifest.csv - auditable row
- qc.md - claim, policy verdicts, failures

## Files

- `gate1.md` - provider, vision stack, delivery, open fields
- `stack.json` - provider, base, auth, endpoints, poll, date
- `source.md` - reference id, URL, date, measured vs inferred
- `layers.md` - layers, transfer and rebuild lines
- `opener.md` - beats, product entry
- `hook.md` - hook type, frame, brand flag
- `banned-words.md` - shell item, reason
- `calibration/` - unit, prompt, seed, cost, variable per round
- `dna.json` - source, hook, emotion, camera, anchor, palette, pacing, falsifier
- `scale.md` - duration, count, angles, hooks, formula
- `queue.md` - variant, angle, hook, duration, provider, status, cap, falsifier
- `gates/` - M1-M4 verdict per cell, failed constraint
- `seams.json` - join index, mean pixel diff
- `assembly/` - EDL, duration, 9:16, 4:5, 1:1
- `manifest.csv` - variant, angle, hook, duration, provider, gate, cost, path
- `qc.md` - claim, policy verdicts, failures

## Cost and time

Hard cap before each confirmed generation. G2 one unit. Cells x cost; cap per call. Failed cell one retry, not batch re-run.

## How this run is judged

Unit is mechanism, not caption. Gate 1 blocks an unrunnable provider so the operator never pays for a stack they cannot run. Gate 2 blocks a batch on an unlocked look: one calibration unit proves the DNA before credits scale. Gate 3 blocks scale that was never agreed: the queue is approved, capped and falsified before any cell renders. The five layers keep measurement and invention apart; the machine gates keep a clean render from being mistaken for a correct one. Skipping a gate is wrong at any scale.

## Stop condition

Stop before G2 until provider named. Before batch until DNA locked and queue approved. Below 80% pass rate, stop batch. Failed M or R line blocks.

## What makes this run fail

- **Batch drifts** - DNA not written. Fix: Lock dna.json G2; paste each cell
- **Provider fails first render** - Adapter assumed. Fix: Probe URL, auth, endpoints before G2
- **Clone is reference+logo** - Surface only. Fix: Five layers; rebuild equation
- **Variants same** - No variable per cell. Fix: Name changed variable each row
- **Seam jumps** - Exit frame not reused. Fix: Chain final frame; measure join
- **Spend past cap** - No cap per cell. Fix: Check cap each generation
- **Batch after failures** - No threshold. Fix: Pass rate per batch; stop below 80%

## Rules

- Evidence before prose. Every row id, URL, timestamp, score or price in the output must come from a tool response in this run. If it did not, label it as an assumption.
- Keep measurement and judgment apart. View counts, cut timings and scores are measured. "This will convert" is a judgment and must read like one.
- No invented links, codes, prices or stock levels. Use only the links the registry returns.
- No artificial cap on shots, scenes or AI-generated cast. Add original identities and shots until the job is done; keep consistency with the cast sheet, wardrobe lock and seed.
- Deduplicate to concepts. Reposts and re-edits of one ad are one data point, not five.
- Ask once for a missing required input. If a tool returns nothing, report the empty result and stop. Never fill a gap with a guess.
- Organic first. Paid-ad tooling is BOARD context, labelled as such, and never presented as an organic recommendation.
- Finish with the deliverable. Do not close with a summary of what you did.

## Shot prompt grammar

- Every shot prompt runs Subject, Action, Camera, Style, Constraints (S-A-C-S-C) in that order: 50-80 words, three sentences, one visual promise. "Cinematic" is not a quality switch - name the light, the lens and the stock.
- Every shot carries a foreground: one blurred object across about a third of the frame, between the lens and the subject, plus one atmospheric layer behind it (fog, dust, steam, a volumetric beam). Say blurred or the model renders it sharp.
- Length matches the file: second-level segments (0s-3s: ...) that add up to the requested duration, because a timeline that does not match makes the model drop the ending. Chain clips by generating the last frame of clip N as the first frame of clip N+1.
- One job per reference image, named in the prompt itself; a reference that already shows the action needs only "strictly follow clip 1". Cap references at 30 images, 10 videos and 10 audio.
- Draft at 480p or the vendor fast tier, gate the first frame, diagnose a failed render in concrete words (material, light path, structure, a negative that did not hold) before re-rolling, and change one variable per retry. Retakes are where the budget goes, not the render.
- Attach the cast sheet to every prompt and never re-describe the character from scratch. Play emotion as body detail, and keep sprints, jumps and rolls out of a single shot - hand the energy to the cut.

## Checks

- R1 G1 answered: provider, stack, delivery, none assumed.
- R2 Adapter fields used verified live, dated.
- R3 No key, token, secret in artifacts.
- R4 Reference id, URL, date; inferred labeled.
- R5 Five layers, transfer and rebuild each.
- R6 Hook one of nine, proven by frame or line.
- R7 Beats ordered; entry measured.
- R8 Shell non-empty; no banned word survives.
- R9 No creator face, voice, brand, claim copied.
- R10 G2 one unit before batch.
- R11 One variable per round, reason logged.
- R12 "lock it" timestamped.
- R13 dna.json parses, filled, falsifier named.
- R14 G3 duration, count, angles, hooks before generation.
- R15 Count = angles x hooks x variants.
- R16 Queue approved before first cell.
- R17 Cell names changed variable.
- R18 Cell has cap and falsifier.
- R19 Batch 10-20.
- R20 Pass rate per batch.
- R21 Below 80% stopped run.
- R22 Failed cell alone, not whole batch.
- R23 M1 casting passed first frame.
- R24 M2 omni passed every clip.
- R25 M3 joins measured; none over 5/255.
- R26 M4 duration exact.
- R27 M5 tokens plus side-by-side.
- R28 M6 claim artifact source/date or removed.
- R29 M7 YES before paid call, cap held.
- R30 M8 policy, likeness, regulated claims.
- R31 Ratios reframed, not re-rendered.
- R32 manifest one row per cell, real files.
- R33 Stranger names structure, different brand.

## Refuse or stop when

- Missing input? Ask once, stop; no invented provider, product, face, claim.
- No generation before G2 lock; no batch before G3 approval.
- No provider named? List adapters, wait; never pick.
- Adapter mismatches docs? Stop, report, no guess.
- Copy creator face/voice/identity? Consent or licensed actor; else original/faceless.
- No proof? Remove claim or invitation to try.
- Cap exceeded? Stop before confirmation, report budget.
- Pass rate below 80%? Stop batch, report drift.
- Regulated? No claim without substantiation.

## Connect

MCP endpoint: `https://sauce-ops.appskins-studio.workers.dev/api/aisle?op=mcp`

Read the full record over MCP with `workflow_get({ id: "recreate" })`.
