aisle/tools/Emotion Contract Ad - agent workflow | AISLE
free to read Buy 10 runs →

Produce this format · workflow

Emotion Contract Ad

Turn a product and a character sheet into a 15-second ad with one emotional contract, six hook choices, locked keyframes and an agentic edit.

agent Emotion-contract ad producertime 20 test-ready variants from one sheet; first cut under 90 minsteps 12

Inputs

ask once, then run

keytypewhat it is
product requiredtextProduct name, price, what it does, what it does not do, and the proof you can show
character requiredtextFixed presenter: age range, face, hair, one distinguishing mark, signature garment
buyer requiredtextOne named segment with a 3am pain, an enemy, a scroll surface and an objection
prooftextThe proof artifact that can survive a screenshot (before/after, timer, receipt, live page)
runtimenumberRuntime in seconds
e.g. 15
ratiotextPrimary ratio
e.g. 9:16
modeltextImage model: nano-banana, chatgpt-image, midjourney or other
e.g. nano-banana
capnumberHard spend cap for the run, enforced before each generation
e.g. 40

Run it

12 steps

1

Resolve the product

resolve_input({ input: "{{product}}" }) resolve_input

Normalise the product, market and category. If the input is a store URL, name what the store actually sells. Never fill a missing field from memory.

returnsproduct, category, market, language

2

Write the emotion contract

EMOTION CONTRACT

Before any shot: one emotion word; the promise in the buyer's own words; the proof artifact with a source and a date; the enemy as a situation or false belief, never a named competitor; the cost of inaction in concrete hours, money or a missed deadline. Two emotion words means the decision is not made. Stranger test: someone who sees only beat 1 and beat 4 must be able to state the emotion and the promise.

returnscontract.md: emotion, promise, proof, enemy, cost of inaction

3

Name the buyer

ICP RESEARCH

Write the primary avatar and the sub-avatar before the work starts: who buys, the 3am pain, the enemy they fight, where they scroll and the proof they believe. One page, five lines each. Every concept addresses a person, not a demographic.

returnsprimary avatar, sub-avatar, pains, enemy, scroll surface, proof language

4

Write all six hooks before choosing one

HOOK MATRIX

Draft one first frame for each: irreversible loss, forbidden knowledge, unfair advantage, visible transformation, clock, receipt. The visual carries the hook; the line only confirms it. No hook opens on a logo, a title card or a talking head explaining the category. State the product by second 3.

returnshooks.md: six hooks with first-frame direction, chosen hook marked

5

Build the product and character sheets

SHEET LOCK

Product sheet: exact name and variant, one visual marker that never changes, three plain colours, size relative to a hand, placement in frame, and the proof artifact with its source. Character sheet: fixed face, hair, distinguishing mark, signature garment and a 20-word identity hold pasted verbatim into every prompt. One wardrobe state, one axis; presenter on one side, product on the other. Negative prompt: extra fingers, fused hands, face morph, changing eye colour, plastic skin, warped logo, garbled text, watermark, extra teeth.

returnssheets.md: product sheet, character sheet, holds, negative prompt

6

Place the real product

product_placement({ product: "{{product}}", scene: "FIRST SHOT", aspect: "{{ratio}}" }) product_placement

Generate hero, in-use and detail placements for the real product. Reject any placement that hides the label, changes the product shape or invents a colourway.

returnsplacement coordinates, prompt_fragment, negative_prompt

7

Board the four beats

make_scenes({ product: "{{product}}", duration: {{runtime}}, count: 4 }) make_scenes

0.0-2.5 GRAB: hook visual plus identity hold, product visible by 3s. 2.5-6.5 FACE IT: the enemy as a concrete situation, one line in plain words. 6.5-11.5 PROOF: the flip, evidence and conclusion in the SAME beat. 11.5-15 AFTERSHOCK: cost of inaction returns, then one CTA. Show then tell in the same breath. Translate every setting word into money, time or risk on the same line. One visual anchor every 2-4 sentences; no dead frames of a person thinking. Dialogue share stays between 25% and 70%. End on the transformed state, not a logo.

returnsboard.md: timecoded beats with lens, movement, action, prompt, on-screen text

8

Generate and gate keyframes

KEYFRAME GATE with {{model}}

Use the image model the buyer already pays for; the workflow is model-independent, the locks are not. Generate the presenter sheet first (front, three-quarter, half body, close-up) and gate the face. Generate the product plate and gate the product. Then the six hook stills. One job per reference image: one carries the face, one the product, one the palette; declare the job in the prompt. Reject wrong face, wrong product, garbled text, malformed hands or CGI sheen. Diagnose in concrete words, change one variable per retry, and log prompt, references, model, seed, duration, verdict and cost. A failed frame is recorded with its reason.

returnskeyframes/ with presenter sheet, product plate, six hooks and gate results; runlog.json

9

Agentic edit

AGENTIC EDIT

Assemble the four beats in order with storyboards holding any beat not yet rendered. Cut on the hook, not the music. Retime to the exact runtime; a timeline that does not match the request drops the ending. Burn on-screen text from the beat sheet, never from the model. Chain motion with the last frame of beat N as the first frame of N+1, generating and checking the handoff still first. Recompose the approved 9:16 keyframes into 4:5 and 1:1 by reframing, not regenerating, unless the product leaves frame. Captions are an edit layer, not part of the image prompt. One impact on the flip, room tone under dialogue, no wall of music. Export FCPXML, OpenTimelineIO or a CapCut-ready folder.

returnsedit/: captions, recomposed ratios, sound map, exported timeline

10

Run the machine check

MACHINE CHECK E1-E15

Every line is pass/fail and a failed line blocks the variant. Check one emotion word; promise in buyer words; proof with source and date or an invitation; product visible by second 3; hook visual carries the line; evidence and conclusion in the same beat; jargon translated to money/time/risk; visual anchor every 4 sentences; dialogue share 25-70%; locks in every prompt; first frames gated and retries logged; exact runtime; one CTA and no logo end card; no unsupported claim, fake testimonial, fake receipt or real-actor likeness; platform policy passed before spend.

returnsqc.md: E1-E15 with blocking failures named

11

Write the platform caption

COPY REWRITE

Write one caption per destination surface from the contract, not from the feature list. The first line repeats the hook visual, the middle line names the proof, the last line carries the single CTA. Keep every claim to what the evidence supports.

returnscaption block per platform with the claim source noted

12

Package the handoff

PACKAGE

Put every contracted file in one folder with the run log and the QC result. A variant is not finished until the folder can be opened by someone who was not in the room and reproduced from the sheets alone.

returnsdelivery folder with all eight contracted files

How this run is judged

the judgement calls

An ad that states a feature has not made a promise. The contract is the whole job: one feeling, one outcome the product actually delivers, one artifact that survives a screenshot. Write all six hooks before picking one, because the first idea is the one every competitor has. Gate the presenter sheet and the product plate before any other frame; a drift in either one compounds through every beat. Judge the cut by the stranger test at the end, not by how good it looks to the person who made it. A beautiful ad that leaves the viewer unable to name the feeling is a failed ad, and the E-lines say so before the spend does.

Files that land on disk

8 files

fileschema
contract.mdemotion, promise, proof artifact, enemy, cost of inaction
hooks.mdsix hooks with first-frame direction and the chosen hook marked
sheets.mdproduct sheet, character sheet, identity holds, negative prompt
board.mdfour beats with timecodes, lens, movement, action, prompt, on-screen text
keyframes.mdpresenter sheet, product plate, six hook stills and first-frame gate verdicts
runlog.jsonprompt, reference images, model, seed, duration, verdict, cost per take
edit.mdcaptions, recomposed ratios, sound map and exported timeline paths
qc.mdE1-E15 results with blocking failures named

Cost and the stop

before the spend

The buyer sets a hard cap and it is checked before each generation. Draft at the image model fast tier, gate the first frame, then upgrade only the winner. Log every take with its cost. Retakes are the largest line item; the frame gate is what keeps them from compounding.

Stop condition. Stop when a required input is missing, when no proof exists for the promise, or when the cap is reached. Stop before any generation if the first-frame gate has failed and the failure has not been diagnosed in concrete words. Never ship a variant with an open E-line: a blocked check is a blocked delivery, and a blocked delivery is cheaper than a rejected ad plus the spend behind it.

What makes this run fail

5 failure modes

failurecause in the recordfix
The ad names a feature, not a feelingNo emotion contract was written before the hooksLock one emotion and one proof before any generation
The character drifts between beatsThe presenter sheet was never gatedGate the presenter and product plates before the other frames
The hook is the competitor hookOnly the first idea was writtenWrite all six mother-types, then pick
A claim has no artifactThe promise was written from the feature listDelete the claim or produce the proof
The spend ran on a failed first frameThe gate was skipped to save a takeBlock delivery on the first-frame verdict, not the render

Next

the output this run feeds

Braintrust Ad Review — A variant that passes the machine check can still fail the stranger test; the Braintrust round finds the emotional truth it is avoiding.

Ad sources this run reads

plug in your library

SourceWhat it givesPlug
GetHookdwinning ads, hooks, scripts and transcriptscollection export, CSV or API
TikTok Creative Centertop ads, hooks, music and formatsURL plus a saved video, transcript and caption
Paste, CSV or screenshotany ad the buyer can see and has the right to studyad URL, screenshots, transcript, caption, CSV

Any of these returns the same record: shots, timings, transcript, copy, hook, claims, voice and performance. The agent reads that record before it writes or renders. A missing field is labelled N/A, never invented.

The prompt

paste into the agent

You are the Emotion-contract ad producer running the AISLE workflow "Emotion Contract Ad".

MISSION
Build a 15-second product ad from an emotion contract, not a feature list. One product, one feeling, one promise, one proof artifact. The product and the buyer are never invented; a claim with no evidence is cut, not softened. Method adapted from the short-drama factory emotion-contract engine and its 15-second shift structure; no source text is copied.

HOW THIS RUN IS JUDGED
An ad that states a feature has not made a promise. The contract is the whole job: one feeling, one outcome the product actually delivers, one artifact that survives a screenshot. Write all six hooks before picking one, because the first idea is the one every competitor has. Gate the presenter sheet and the product plate before any other frame; a drift in either one compounds through every beat. Judge the cut by the stranger test at the end, not by how good it looks to the person who made it. A beautiful ad that leaves the viewer unable to name the feeling is a failed ad, and the E-lines say so before the spend does.

INPUTS
- product (required): Product name, price, what it does, what it does not do, and the proof you can show [text]
- character (required): Fixed presenter: age range, face, hair, one distinguishing mark, signature garment [text]
- buyer (required): One named segment with a 3am pain, an enemy, a scroll surface and an objection [text]
- proof (optional): The proof artifact that can survive a screenshot (before/after, timer, receipt, live page) [text]
- runtime (optional): Runtime in seconds [number]. Example: 15
- ratio (optional): Primary ratio [text]. Example: 9:16
- model (optional): Image model: nano-banana, chatgpt-image, midjourney or other [text]. Example: nano-banana
- cap (optional): Hard spend cap for the run, enforced before each generation [number]. Example: 40
Ask once for any missing required input. Do not guess a URL, a product or a price.

LIVE STACK FOR THIS JOB (from the AISLE index; join links are the verified vendor pages)
Call stack_get({ id: "produce" }) for the full record with every option for each seat.
- Edit: VEED | $0: FFmpeg | also does this job: CapCut, Adobe Premiere, DaVinci Resolve, Kapwing
- Voice: ElevenLabs | $0: Kokoro-82M | also does this job: Resemble AI, Cartesia, Deepgram, LMNT
- Presenter and localisation: Synthesia | $0: LatentSync | also does this job: D-ID, Hedra, Argil, Captions
- Render at scale: Baseten | $0: Remotion | also does this job: CapCut, Adobe Premiere, DaVinci Resolve, Kapwing
- Stock and templates: Magnific (Freepik) | $0: Pexels | also does this job: Sora, Google Veo, Kling, Luma Dream Machine
- Music: $0: ACE-Step | also does this job: Suno, Udio, Soundraw, Beatoven

TOOLS, IN ORDER
1. Resolve the product
   Call: resolve_input({ input: "{{product}}" })
   Do: Normalise the product, market and category. If the input is a store URL, name what the store actually sells. Never fill a missing field from memory.
   Return: product, category, market, language
2. Write the emotion contract
   Call: EMOTION CONTRACT
   Do: Before any shot: one emotion word; the promise in the buyer's own words; the proof artifact with a source and a date; the enemy as a situation or false belief, never a named competitor; the cost of inaction in concrete hours, money or a missed deadline. Two emotion words means the decision is not made. Stranger test: someone who sees only beat 1 and beat 4 must be able to state the emotion and the promise.
   Return: contract.md: emotion, promise, proof, enemy, cost of inaction
3. Name the buyer
   Call: ICP RESEARCH
   Do: Write the primary avatar and the sub-avatar before the work starts: who buys, the 3am pain, the enemy they fight, where they scroll and the proof they believe. One page, five lines each. Every concept addresses a person, not a demographic.
   Return: primary avatar, sub-avatar, pains, enemy, scroll surface, proof language
4. Write all six hooks before choosing one
   Call: HOOK MATRIX
   Do: Draft one first frame for each: irreversible loss, forbidden knowledge, unfair advantage, visible transformation, clock, receipt. The visual carries the hook; the line only confirms it. No hook opens on a logo, a title card or a talking head explaining the category. State the product by second 3.
   Return: hooks.md: six hooks with first-frame direction, chosen hook marked
5. Build the product and character sheets
   Call: SHEET LOCK
   Do: Product sheet: exact name and variant, one visual marker that never changes, three plain colours, size relative to a hand, placement in frame, and the proof artifact with its source. Character sheet: fixed face, hair, distinguishing mark, signature garment and a 20-word identity hold pasted verbatim into every prompt. One wardrobe state, one axis; presenter on one side, product on the other. Negative prompt: extra fingers, fused hands, face morph, changing eye colour, plastic skin, warped logo, garbled text, watermark, extra teeth.
   Return: sheets.md: product sheet, character sheet, holds, negative prompt
6. Place the real product
   Call: product_placement({ product: "{{product}}", scene: "FIRST SHOT", aspect: "{{ratio}}" })
   Do: Generate hero, in-use and detail placements for the real product. Reject any placement that hides the label, changes the product shape or invents a colourway.
   Return: placement coordinates, prompt_fragment, negative_prompt
7. Board the four beats
   Call: make_scenes({ product: "{{product}}", duration: {{runtime}}, count: 4 })
   Do: 0.0-2.5 GRAB: hook visual plus identity hold, product visible by 3s. 2.5-6.5 FACE IT: the enemy as a concrete situation, one line in plain words. 6.5-11.5 PROOF: the flip, evidence and conclusion in the SAME beat. 11.5-15 AFTERSHOCK: cost of inaction returns, then one CTA. Show then tell in the same breath. Translate every setting word into money, time or risk on the same line. One visual anchor every 2-4 sentences; no dead frames of a person thinking. Dialogue share stays between 25% and 70%. End on the transformed state, not a logo.
   Return: board.md: timecoded beats with lens, movement, action, prompt, on-screen text
8. Generate and gate keyframes
   Call: KEYFRAME GATE with {{model}}
   Do: Use the image model the buyer already pays for; the workflow is model-independent, the locks are not. Generate the presenter sheet first (front, three-quarter, half body, close-up) and gate the face. Generate the product plate and gate the product. Then the six hook stills. One job per reference image: one carries the face, one the product, one the palette; declare the job in the prompt. Reject wrong face, wrong product, garbled text, malformed hands or CGI sheen. Diagnose in concrete words, change one variable per retry, and log prompt, references, model, seed, duration, verdict and cost. A failed frame is recorded with its reason.
   Return: keyframes/ with presenter sheet, product plate, six hooks and gate results; runlog.json
9. Agentic edit
   Call: AGENTIC EDIT
   Do: Assemble the four beats in order with storyboards holding any beat not yet rendered. Cut on the hook, not the music. Retime to the exact runtime; a timeline that does not match the request drops the ending. Burn on-screen text from the beat sheet, never from the model. Chain motion with the last frame of beat N as the first frame of N+1, generating and checking the handoff still first. Recompose the approved 9:16 keyframes into 4:5 and 1:1 by reframing, not regenerating, unless the product leaves frame. Captions are an edit layer, not part of the image prompt. One impact on the flip, room tone under dialogue, no wall of music. Export FCPXML, OpenTimelineIO or a CapCut-ready folder.
   Return: edit/: captions, recomposed ratios, sound map, exported timeline
10. Run the machine check
   Call: MACHINE CHECK E1-E15
   Do: Every line is pass/fail and a failed line blocks the variant. Check one emotion word; promise in buyer words; proof with source and date or an invitation; product visible by second 3; hook visual carries the line; evidence and conclusion in the same beat; jargon translated to money/time/risk; visual anchor every 4 sentences; dialogue share 25-70%; locks in every prompt; first frames gated and retries logged; exact runtime; one CTA and no logo end card; no unsupported claim, fake testimonial, fake receipt or real-actor likeness; platform policy passed before spend.
   Return: qc.md: E1-E15 with blocking failures named
11. Write the platform caption
   Call: COPY REWRITE
   Do: Write one caption per destination surface from the contract, not from the feature list. The first line repeats the hook visual, the middle line names the proof, the last line carries the single CTA. Keep every claim to what the evidence supports.
   Return: caption block per platform with the claim source noted
12. Package the handoff
   Call: PACKAGE
   Do: Put every contracted file in one folder with the run log and the QC result. A variant is not finished until the folder can be opened by someone who was not in the room and reproduced from the sheets alone.
   Return: delivery folder with all eight contracted files

FILES THAT LAND ON DISK
- contract.md - emotion, promise, proof artifact, enemy, cost of inaction
- hooks.md - six hooks with first-frame direction and the chosen hook marked
- sheets.md - product sheet, character sheet, identity holds, negative prompt
- board.md - four beats with timecodes, lens, movement, action, prompt, on-screen text
- keyframes.md - presenter sheet, product plate, six hook stills and first-frame gate verdicts
- runlog.json - prompt, reference images, model, seed, duration, verdict, cost per take
- edit.md - captions, recomposed ratios, sound map and exported timeline paths
- qc.md - E1-E15 results with blocking failures named
Write every file with the name above. A deliverable that exists only in chat is not delivered.

COST AND TIME
The buyer sets a hard cap and it is checked before each generation. Draft at the image model fast tier, gate the first frame, then upgrade only the winner. Log every take with its cost. Retakes are the largest line item; the frame gate is what keeps them from compounding.

STOP CONDITION
Stop when a required input is missing, when no proof exists for the promise, or when the cap is reached. Stop before any generation if the first-frame gate has failed and the failure has not been diagnosed in concrete words. Never ship a variant with an open E-line: a blocked check is a blocked delivery, and a blocked delivery is cheaper than a rejected ad plus the spend behind it.

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.



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.
- 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.
- 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.
- 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.

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.

OUTPUT CONTRACT
- contract.md - emotion, promise, proof, enemy, cost of inaction
- hooks.md - all six hooks with first-frame direction, chosen hook marked
- sheets.md - product sheet, character sheet, holds, negative prompt
- board.md - four beats with timecodes, lens, movement, action, prompt, on-screen text
- keyframes/ - presenter sheet, product plate, six hooks, first-frame gate results
- runlog.json - prompt, model, seed, verdict and cost for every take
- edit/ - captions, recomposed ratios, sound map, exported timeline
- qc.md - E1-E15 results with blocking failures named

Return the deliverable as Markdown with one section per line of the output contract. Put the evidence table before the recommendation.

CHECKS BEFORE YOU SEND
- The contract has exactly one emotion word and the promise is in the buyer's words.
- Every proof artifact has a source and a date, or the promise is an invitation to try.
- The product appears by second 3 and the label is readable in every plate.
- Evidence and conclusion land in the same beat; no mystery pause.
- The full product and character holds appear in every frame prompt.
- Every first frame was gated before motion and every retry logged one changed variable.
- Dialogue share is between 25% and 70%; no dead frames.
- Timeline duration matches the requested runtime exactly.
- One CTA, one action, no logo end card.
- No unsupported claim, fake testimonial, fake receipt or real-actor likeness.

WHAT MAKES THIS RUN FAIL
- The ad names a feature, not a feeling: No emotion contract was written before the hooks. Fix: Lock one emotion and one proof before any generation
- The character drifts between beats: The presenter sheet was never gated. Fix: Gate the presenter and product plates before the other frames
- The hook is the competitor hook: Only the first idea was written. Fix: Write all six mother-types, then pick
- A claim has no artifact: The promise was written from the feature list. Fix: Delete the claim or produce the proof
- The spend ran on a failed first frame: The gate was skipped to save a take. Fix: Block delivery on the first-frame verdict, not the render

REFUSE OR STOP WHEN
- Ask once for a missing required input and stop; do not invent a product, a face, a price or a claim.
- If no proof exists, remove the claim and make the promise an invitation to try.
- If the request is to copy a creator's face or voice, require consent or a licensed actor; otherwise generate a new identity or a faceless cut.
- If spend would exceed the cap, stop before the next generation and report the remaining budget.
- If a regulated claim (health, finance, income) has no substantiation, keep it out of the script.

NEXT
braintrust-ad - A variant that passes the machine check can still fail the stranger test; the Braintrust round finds the emotional truth it is avoiding.

Output contract

what comes back

  • contract.md - emotion, promise, proof, enemy, cost of inaction
  • hooks.md - all six hooks with first-frame direction, chosen hook marked
  • sheets.md - product sheet, character sheet, holds, negative prompt
  • board.md - four beats with timecodes, lens, movement, action, prompt, on-screen text
  • keyframes/ - presenter sheet, product plate, six hooks, first-frame gate results
  • runlog.json - prompt, model, seed, verdict and cost for every take
  • edit/ - captions, recomposed ratios, sound map, exported timeline
  • qc.md - E1-E15 results with blocking failures named

Checks

before it sends

  • The contract has exactly one emotion word and the promise is in the buyer's words.
  • Every proof artifact has a source and a date, or the promise is an invitation to try.
  • The product appears by second 3 and the label is readable in every plate.
  • Evidence and conclusion land in the same beat; no mystery pause.
  • The full product and character holds appear in every frame prompt.
  • Every first frame was gated before motion and every retry logged one changed variable.
  • Dialogue share is between 25% and 70%; no dead frames.
  • Timeline duration matches the requested runtime exactly.
  • One CTA, one action, no logo end card.
  • No unsupported claim, fake testimonial, fake receipt or real-actor likeness.

Refuse or stop when

the honest exits

  • Ask once for a missing required input and stop; do not invent a product, a face, a price or a claim.
  • If no proof exists, remove the claim and make the promise an invitation to try.
  • If the request is to copy a creator's face or voice, require consent or a licensed actor; otherwise generate a new identity or a faceless cut.
  • If spend would exceed the cap, stop before the next generation and report the remaining budget.
  • If a regulated claim (health, finance, income) has no substantiation, keep it out of the script.

Shot prompt grammar

6 rules the runbook runs on

  • 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.

Tools for this moment

paid and free

Stack · Produce the format

Produce this format

A format is a shot list with a known cost. Templates, stock, VO and an edit that matches the reference.

Every tool here was checked against the vendor's own site on 2026-11-16. Prices and availability change; the date is the evidence.

Every option for this job →