aisle/tools/Product Film — shot board, prompts and agent pack | AISLE
free to read Buy 10 runs →

Studio · Product · 9:16 + 16:9 · styleframe-first

Product Film

A styleframe-first product film: lock palette, type and motion in cheap stills, map each feature to one shot, then render deterministically and gate every shot. The product is captured, not imagined; text is typeset, not drawn; and the cut is finished only on the shots that survived the styleframes.

8 shots / 30s film core boardcast to script13 moodboards9 variantsNo ceilingNo ceiling

The whole line, one pack

story to stitch

01 · Product truthCapture the product page: images, features, materials, dimensions, label artwork. Record the URL and the date.
02 · LookExtract three colour swatches, two type styles and one motion rule as tokens. Styleframes before motion.
03 · Product lockHero angle, label, cap, liquid, packaging: one card pasted into every prompt. No invented variant, no borrowed SKU.
04 · BoardEvery shot gets a timecode, lens, movement, product state, on-screen text and prompt. Map each feature to one shot.
05 · GenerateMacro keyframes, pour and spray, image-to-video and SFX. Three takes; keep the one where the label stays legible.
06 · ConsistencySame pack, label, colour and light in every shot. Paste the lock, reuse the seed, check the label before you cut.
07 · SoundRoom tone, pour and spray foley, one music bed. No fake testimonial and no invented claim.
08 · CutTimeline, grade, captions, delivery in 16:9, 9:16 and 1:1. The cut is the product.
09 · RemasterProve the cut on a cheap model, then finish only the used clips at their used lengths on the premium model.

The board below is the core, not a ceiling; extend it until the story lands.

What you get

11 deliverables

  • Product understanding brief from the real page: features, materials, proportions, design tokens
  • Styleframes locking palette, type, light and camera
  • Feature-to-shot map, one feature per shot
  • Deterministic shot renders from a fixed seed
  • Two QA frames per shot, checked against the product reference
  • Typeset text layer with real fonts
  • Two cutdowns: 30 s master and 15 s paid
  • Delivery manifest with ratios, codecs and loudness
  • Token sheet: three colours, two type styles, one motion rule, written as values
  • Acceptance frames per shot: the first and last frame of every move
  • Font licence record for every typeface used in the edit

World and look

paste before anything

WorldYOUR PRODUCT - real captures, design tokens and the feature list from the product page. The world is the product's own material language: brushed metal, matte ceramic, woven textile. No invented brand, no invented feature.
PaletteYOUR TOKENS - extract colour, type and motion from the product; do not invent a brand. Three swatches, two type styles, one motion rule. The background is a chosen neutral, not a gradient.
Negativeextra fingers, fused hands, face morph, changing eye colour, plastic skin, warped jewellery, floating earrings, captions, watermark, logo, extra teeth

Locks

location, props, wardrobe, axis

productProportions, silhouette, material, label and finish from the real capture. Any deviation in proportion or label is a failed take, not a style choice.
surfaceOne surface per scene: stone, linen, water, skin. State the material and its finish in the prompt; never let the model invent a reflective floor.
lightOne key, one fill, one edge. Name the direction and the quality in every prompt. A new light setup is a new scene.
typeReal fonts, typeset in the edit. The model never draws a word. Type scale, tracking and safe area are fixed in the styleframe.
motionOne motion rule for the film: either the camera moves and the product is still, or the product moves and the camera is locked. Never both.
scaleReal dimensions from the product page, stated as a ratio against a known object: a hand, a coin, a card. Scale is the fastest way to make a real product look fake.
editingCut on the beat or on a motion match. One transition type for the whole film: the cut. No whip, no zoom blur, no glitch.
handThe hand in frame is part of the cast: one skin tone, one sleeve, short nails, always the same side of the product. A second hand is a second card.

Every lock is pasted into the prompt for every shot that touches it. A lock that is paraphrased is not a lock.

Cast locks

0 identities

No cast. This pack is product, archive or graphic-led.

Shot prompt grammar

16 rules every prompt on this board obeys

  • Order: Subject, Action, Camera, Style, Constraints (S-A-C-S-C). Subject first so the model binds the identity before it reads motion.
  • Actions stay slow, soft and continuous. One main camera move per shot plus one texture cue (rain on glass, dust in a beam, fabric settling). Two moves in one prompt cancel each other.
  • 50-80 words, three sentences, one visual promise. "Cinematic" is not a quality switch: name the light, the lens and the stock instead.
  • Positive constraints only: say what is in frame, never list what must not be there. Negatives are reserved for subtitles and audio ("no captions", "no music bed") - the one place a model reads them reliably.
  • Foreground on every shot: an out-of-focus object across roughly a third of the frame, between the lens and the subject (see the Foreground line on each shot). Say "blurred" or the model renders it sharp. Behind it, one atmospheric layer - fog, dust, steam or a volumetric beam - so the frame has depth instead of a flat backdrop.
  • Length matches the file. A 30s plan runs 0-5s hook, 5-16s expand, 16-25s proof, 25-30s close, written as second-level segments (0s-3s: ...). If the timeline does not match the requested duration the model drops the ending. On a model that ignores seconds, fall back to shot numbers; on 2.5-class models seconds are first-class.
  • Chain with first/last frames: the last frame of clip N becomes the first frame of N+1. Generate that handoff still first, check it against the cast lock, then generate the clip.
  • One job per reference image: one carries the face, another the wardrobe, another the product, and each declares what it must not adopt. Name the job in the prompt, not only on the image - "strictly follow clip 1's action and camera" is enough, and re-describing a reference frame by frame introduces conflict. Stay inside the model cap (Seedance 2.5: 30 images, 10 videos, 10 audio).
  • Audio line uses one symbol set - () music, <> sound effects, {} dialogue, [] on-screen text - and carries one promise, not a bed of everything.
  • Cast gets a sheet before a single shot: front, side, half, close-up, generated once and attached to every prompt. Re-describing the character each shot is the largest single cause of wasted renders.
  • Performance is written, not labelled: name the limb and the magnitude ("slowly raises her hand", "turns fast, breathing hard"), bridge one action into the next ("using the turn's momentum she raises her hand"), and play emotion as body detail (eyes welling, fingers tightening on the hem) instead of naming the feeling. Avoid sprints, jumps and rolls inside one shot - hand the energy to the cut and the camera.
  • Light gets its own sentence: direction, quality (hard or soft), contrast. Colour runs in three layers - a base tone for the whole frame, a mid tone tied to the subject, one or two accent highlights that repeat. Then delete the word cinematic.
  • Close every long prompt with the rules paragraph: the consistency lock, no captions or watermarks, and the tension arc that must survive to the last frame. It is the red line the model does not cross.
  • Debug at draft resolution. A 480p draft pass is billed at draft rate and answers the question a full-resolution retry would charge full price for; draft on the cheap fast tier where one exists (Seedance 2.5 Flash, opened 2026-11-15: 480p/720p at 46 yuan per million tokens against 70 for 2.5, about twice as fast), and upgrade only the winner. Retakes eat most of a budget - roughly seven of ten credits die on a bad take - so previs the complex shot as a white model first (7-10 retries down to 2-3), shoot 3-5s pieces and stitch, and change one variable per retry.
  • Gate the first frame before you scale: wrong face or product, garbled text, malformed hands, CGI look. A failed render gets diagnosed in concrete words first - material, light path, aerial perspective, overlapping structure, a negative that did not hold - then repaired with those words. A blind re-roll with a new seed doubles the bill; only after the repair fails do you change the seed.
  • Read the reference post before you write: pull its transcript and overview frame sheet locally so line timing and cut rhythm come from the real post, not a guess.

The same block is baked into the SKILL.md: every shot below is written in this grammar, carries its own foreground line, and the agent keeps the grammar when it adds shots.

Skills this pack loads

5 skills, in order

SkillJob in this run
copywritinghooks, headlines and page copy
ad-creativead copy at scale
imagekeyframes, model sheets and moodboards
videoshots, motion, editing and the cut
socialorganic and platform format

The SKILL.md installs these in order. Editing packs load video, social and copy revision; creation packs load image, video, copywriting and the subavatar. A pack never loads a skill it does not use.

Ad sources this pack reads

plug in your library

SourceWhat it givesPlug
Brandsearchbrand ads across Meta, Google and TikTok, plus landing pages and copy historyboard export, URL or API
Foreplayswipe file, ad copy, video downloads and timestampsboard export or API
Meta Ad Librarylive ads, copy, start dates and platformsad URL plus a saved screenshot, transcript and caption
Paste, CSV or screenshotany ad the buyer can see and has the right to studyad URL, screenshots, transcript, caption, CSV

If the buyer names a library, the agent uses its export or API. If not, it asks for one export, URL, screenshot, transcript or CSV. Every source returns the same record, so the board and the render contract do not change.

Any model, any seat

swap on a rate limit

Claude, ChatGPT, DeepSeek, Grok, Kimi, Muse, Gemini, Qwen and Seedance can each hold any seat: read the source, write the board, hold the brand voice, run QC. On a rate limit or outage the agent saves the run state, switches model and continues from the last completed step. It never restarts the run and never re-charges a finished take. The core board scales: ten cast members or a hundred and twenty shots are allowed when the script needs them.

The shot board

10 core shots · full prompts

SHOT 01
0.0-2.5s
macro
slow push

Hero — Extreme close-up of the real product. One edge, one material, one highlight.

macro · slow push · 0.0-2.5s

Macro of [PRODUCT] from the real capture, extracted design tokens, one key from camera left, blurred foreground across a third of the frame, vertical 9:16, no text

foregroundthe soft edge of a shoulder across the right third, out of focus, between the lens and the subject

audioRiser into impact.

SHOT 02
2.5-6.0s
50mm
locked

Whole object — The full product, clean and exact, at correct proportion.

50mm · locked · 2.5-6.0s

Front three-quarter of [PRODUCT], fixed proportions from the capture, one soft box, 50mm, locked, [SURFACE], vertical 9:16, no text

foregroundthe soft edge of a shoulder across the right third, out of focus, between the lens and the subject

audioSettled impact tail.

line[name the product once]

SHOT 03
6.0-10.0s
35mm
lateral track

Feature one — One real feature, one motion. The camera moves, the product stays still.

35mm · lateral track · 6.0-10.0s

Feature detail from the capture sheet, real design tokens, 35mm, lateral track, [LIGHT], vertical 9:16, no text

foregrounda hanging fabric edge, blurred across the top third, out of focus, between the lens and the subject

audioUI or material sound.

on screen[feature name]

line[the feature as a benefit, one line]

SHOT 04
10.0-14.0s
50mm
slow push

Feature two — A second feature proves the first claim. Same light logic, new angle.

50mm · slow push · 10.0-14.0s

Second feature detail, same light logic, 50mm, slow push, [SURFACE], vertical 9:16, no text

foregrounda blurred strand of hair across the left third, out of focus, between the lens and the subject

audioSecond sound, same family.

on screen[feature two]

line[the proof in one line]

SHOT 05
14.0-18.0s
85mm
locked

Material — Finish and material under a moving light. The texture does the talking.

85mm · locked · 14.0-18.0s

Material close of [PRODUCT], light sweep, one texture cue, 85mm, locked, vertical 9:16, no text

foregrounda soft foreground shape thrown out of focus across the bottom third, out of focus, between the lens and the subject

audioMaterial sound, close.

SHOT 06
18.0-22.0s
35mm
orbit

In use — The product in a real hand, real scale. The hand enters from frame left.

35mm · orbit · 18.0-22.0s

In-use shot of [PRODUCT], real hand for scale, practical light, 35mm, orbit, vertical 9:16, no text

foregrounda blurred strand of hair across the left third, out of focus, between the lens and the subject

audioReal use sound.

line[what changes for the user]

SHOT 07
22.0-26.5s
24mm
pull back

Comparison — The product beside the thing it replaces. Both real, same light.

24mm · pull back · 22.0-26.5s

Two-shot of [PRODUCT] and [OLD OBJECT], fixed light, 24mm, pull back, [SURFACE], vertical 9:16, no text

foregroundthe soft edge of a shoulder across the right third, out of focus, between the lens and the subject

audioComparison hit.

on screen[the difference in 4 words]

line[the comparison, no exaggeration]

SHOT 08
26.5-30.0s
50mm
locked

Front freeze — Clean front, name and CTA typeset in the edit. The product holds still for one second.

50mm · locked · 26.5-30.0s

Front freeze of [PRODUCT], clean [SURFACE], fixed key, 50mm, locked, vertical 9:16, text added in the edit

foregroundthe soft edge of a shoulder across the right third, out of focus, between the lens and the subject

audioResolve chord.

on screen[name, price if real, CTA]

SHOT 09
30.0-33.0s
50mm
locked

End card — The same front frame with the offer line and the URL, typeset.

50mm · locked · 30.0-33.0s

Same front freeze, clean background, space in the lower third, 50mm, locked, vertical 9:16

foregrounda passing guard rail, smeared by motion across the right third, out of focus, between the lens and the subject

audioTail, no new element.

on screen[offer and URL]

SHOT 10
33.0-35.0s
macro
slow pull

Loop — A macro detail that rhymes with shot one, so the film loops cleanly.

macro · slow pull · 33.0-35.0s

Macro detail of [PRODUCT], same key as shot 01, 35mm, slow pull, vertical 9:16, no text

foregroundthe soft edge of a shoulder across the right third, out of focus, between the lens and the subject

audioRiser tail.

Core board, not a cap. Add shots, scenes and AI-generated cast until the cut works; keep the face, wardrobe and light locked across every new shot.

Beat sheet

8 beats

TimeBeatJob
0-2sHeroProduct as treasure. Macro detail, one light, no context yet.
2-6sWhole objectThe full silhouette at correct proportion. The audience learns the shape before the features.
6-10sFeature oneOne real feature, one motion, shown as cause and effect.
10-14sFeature twoA second feature proves the first claim. Same light logic.
14-18sMaterialFinish and material under a moving light. Texture is the proof.
18-22sIn useThe product in a real hand at real scale. Nothing floats.
22-26sComparisonThe product beside the thing it replaces. The difference is visible, not stated.
26-30sFront freezeClean front, name and CTA typeset. One frame the audience can remember.

Moodboard prompts

13 images

Styleframe · 16:9Two or three stills locking palette, type, light and camera before any motion. The chosen frame becomes the reference for every shot.
Motion sheet · 16:9Energy and tone axes mapped from the real brand, with one motion rule stated in words.
Material board · 1:1Three material swatches from the product capture: finish, texture, edge. Used to keep the model honest.
Type lock · 9:16The type system on a clean frame: scale, tracking, safe area, one accent. Typeset, never generated.
QA frame pair · 16:9Two fixed acceptance frames per shot: one at the start of the move, one at the end, checked against the product capture.
Token sheet · 16:9Three swatches with hex values, two type styles with weights, one motion rule in a sentence. This is the contract the renderer reads, not a mood.
Silhouette check · 1:1The product reduced to a black silhouette at the correct proportion, beside the real capture silhouette. If they do not overlay, the render is wrong before it starts.
Light study · 16:9The same product under the key, the fill and the edge, one frame each, so the chosen setup is a decision rather than a prompt adjective.
Cut map · 16:9Every shot on one strip with its beat, its length and its motion rule. The paid cut is marked on the same strip, not built separately.
Feature map · 16:9Every feature on the product, numbered, with the shot that carries it beside it. A feature without a shot is a feature the film does not claim.
Loop frame · 9:16The last frame beside the first, checked so the film loops without a visible seam on a feed.
Hand scale · 1:1The product beside a real hand and a coin, shot at the same distance. Scale is checked against two familiar objects, not one.
Ratio pair · 16:9The same shot framed for 16:9 and 9:16 side by side, with the product's safe area marked. The vertical frame is composed, not cropped.

Hooks and variants

9 openers

  • Feature-led: open on the hardest feature and end on the whole object.
  • Material-led: no features, only texture and light, for a premium price point.
  • Comparison-led: open on the old object and reveal the product as the replacement.
  • Silent cut: type and product only, built for autoplay.
  • Founder cut: the same board with a real hand and a real voice.
  • 16:9 cinema cut and 9:16 paid cut from the same board, re-framed not cropped.
  • One-shot film: a single continuous move around the product, no cuts, for a premium placement.
  • Silent loop: six seconds, product and type only, built for a muted feed.
  • Detail-only cut: no whole object until the final frame.

Sound and music

direction, not a playlist

Design the film as one arc: a riser into the hero, a clean impact, then material and UI sounds that belong to the product. Music is minimal and low; the product sounds carry the proof. No stock whoosh on a cut that does not need one. Loudness -16 LUFS for web, -14 for paid social, true peak -1 dB. Build the sound from the product: the lid, the clasp, the pour, the click. Those recordings are the proof the object is real. Music is one bed with a single lift at the feature that matters; if the bed is doing the persuading, the shot is not.

QC gate

13 checks, pass or fail

CheckPassFail
ProportionSilhouette and scale match the product captureA stretched or resized product: discard the take
LabelLabel, logo and finish are the real onesRedrawn text: rebuild from the capture
ColourProduct colour matches the swatch under the stated lightA hue shift: re-render with the token
TextAll text is typeset with real fonts and reads at 390 pxModel-drawn glyphs: delete the layer
MotionOne motion rule holds for the whole filmCamera and product moving together: re-board
MaterialTexture reads as the real material, not a generic glossy surfacePlastic sheen: change the light, not the prompt adjective
SeamsCuts land on the beat or on a motion matchA visible jump: trim and re-join
LoudnessMaster hits the target with no clippingOver: re-mix
DeliveryBoth ratios, captions where spoken, AI label where syntheticMissing asset: block delivery
SilhouetteThe rendered silhouette overlays the capture silhouetteA mismatch: fix proportion before lighting
FontEvery typeface is licensed for commercial use and recordedAn unlicensed face: replace it before delivery
CutOne transition type, cuts on beat or motion matchA stray transition: remove it
SurfaceThe surface material matches the stated finish in the promptA generic glossy floor: change the surface, not the adjective

What would make this fail

11 known failures

SymptomCauseFix
Product looks like a different objectPrompt described a category instead of the capturePaste the capture and the proportion lock, regenerate
Text is garbledThe model was asked to draw typeRemove text from the prompt, typeset in the edit
Film feels like a slideshowEvery shot is locked and the same distanceAdd one move per feature and vary the lens
Colour drifts between shotsThe palette was paraphrasedUse the token names verbatim and re-render
Product floatsNo surface or hand for scaleAdd the surface lock and a real hand
Cuts feel randomNo beat mapMark the music beats before the edit and cut on them
Paid cut loses the productThe 9:16 was cropped from 16:9Re-frame the board for 9:16 and re-render the used shots
Product reads as a renderScale was described in adjectives instead of a ratioAdd the ratio lock and a real hand for reference
Film feels cheapToo many transitions and no sound from the objectCut with the cut, and record the product's own sounds
Paid cut loses the labelThe 9:16 was cropped from the centre of 16:9Re-frame the board for 9:16 and re-render the used shots
Feature shots look alikeEvery feature was shot at the same distanceVary lens and distance per feature, and give each one its own motion

Agent runbook

18 steps

Produce a product film from real captures and design tokens. Settle the direction in cheap styleframes before any motion, map one feature to one shot, render deterministically, and never let the model draw a word. The product is the reference; the film is a reading of it.

1

Capture the product page: images, feature list, materials, dimensions, label artwork. Record the URL and the date.

2

Extract three colour swatches, two type styles and one motion rule. Write them as tokens, not adjectives.

3

Make two or three styleframes and pick one. Nothing moves until the styleframe is approved.

4

Map each feature to one shot and write the board: lens, move, action, on-screen text and line.

5

Price the board and set a cap. Cut shots before cutting quality if the quote is over.

6

Render the first shot from a fixed seed and compare it to the capture. Only then render the rest.

7

Check two acceptance frames per shot: start of move and end of move.

8

Assemble with temp music, then replace the music and cut on the beats.

9

Typeset every word in the edit with the locked fonts and safe area.

10

Re-render only failed shots, at their used lengths. Never re-render the whole film to fix one frame.

11

Export 16:9 and 9:16 from the same board, with captions and the AI label.

12

Write the delivery manifest: ratios, codecs, loudness, fonts, seeds and the accepted takes.

13

Write the token sheet as values: hex colours, font names and weights, one motion rule. Nothing in the film may use a value that is not on the sheet.

14

Overlay the rendered silhouette on the capture silhouette before lighting a single frame.

15

Record the product's own sounds before the edit and build the mix from them.

16

Mark the paid cut on the same strip as the master and re-frame, never crop.

17

Number every feature and name the shot that carries it before the board is priced.

18

Check the loop frame against the first frame before the final export.

The prompt

paste into Claude or Codex

You are the production agent for the AISLE playbook "Product Film".

MISSION
Produce a product film from real captures and design tokens. Settle the direction in cheap styleframes before any motion, map one feature to one shot, render deterministically, and never let the model draw a word. The product is the reference; the film is a reading of it.

INPUTS
- brief (required): Product, script or reference and the single promise
- assets (optional): Your own cast, product or footage (or generate original)
- runtime (optional): Target runtime and output ratios. Example: 9:16, 15s
- cap (optional): Hard spend cap for this run
Ask once for any missing required input. Do not invent a product, a face or a claim.

PRODUCTION LINE (run in this order; this is the whole job)
01. Product truth: Capture the product page: images, features, materials, dimensions, label artwork. Record the URL and the date.
02. Look: Extract three colour swatches, two type styles and one motion rule as tokens. Styleframes before motion.
03. Product lock: Hero angle, label, cap, liquid, packaging: one card pasted into every prompt. No invented variant, no borrowed SKU.
04. Board: Every shot gets a timecode, lens, movement, product state, on-screen text and prompt. Map each feature to one shot.
05. Generate: Macro keyframes, pour and spray, image-to-video and SFX. Three takes; keep the one where the label stays legible.
06. Consistency: Same pack, label, colour and light in every shot. Paste the lock, reuse the seed, check the label before you cut.
07. Sound: Room tone, pour and spray foley, one music bed. No fake testimonial and no invented claim.
08. Cut: Timeline, grade, captions, delivery in 16:9, 9:16 and 1:1. The cut is the product.
09. Remaster: Prove the cut on a cheap model, then finish only the used clips at their used lengths on the premium model.
Receipt for every take: model, seconds, cost. A budget cap per film that is enforced. No frame generated before it is priced.

WORLD
YOUR PRODUCT - real captures, design tokens and the feature list from the product page. The world is the product's own material language: brushed metal, matte ceramic, woven textile. No invented brand, no invented feature.
Palette: YOUR TOKENS - extract colour, type and motion from the product; do not invent a brand. Three swatches, two type styles, one motion rule. The background is a chosen neutral, not a gradient.
Negative prompt: extra fingers, fused hands, face morph, changing eye colour, plastic skin, warped jewellery, floating earrings, captions, watermark, logo, extra teeth

CAST LOCKS (paste the full lock into every prompt)
- No cast. This is a product, archive or graphic-led film.

LOCKS (location, props, wardrobe, axis)
- product: Proportions, silhouette, material, label and finish from the real capture. Any deviation in proportion or label is a failed take, not a style choice.
- surface: One surface per scene: stone, linen, water, skin. State the material and its finish in the prompt; never let the model invent a reflective floor.
- light: One key, one fill, one edge. Name the direction and the quality in every prompt. A new light setup is a new scene.
- type: Real fonts, typeset in the edit. The model never draws a word. Type scale, tracking and safe area are fixed in the styleframe.
- motion: One motion rule for the film: either the camera moves and the product is still, or the product moves and the camera is locked. Never both.
- scale: Real dimensions from the product page, stated as a ratio against a known object: a hand, a coin, a card. Scale is the fastest way to make a real product look fake.
- editing: Cut on the beat or on a motion match. One transition type for the whole film: the cut. No whip, no zoom blur, no glitch.
- hand: The hand in frame is part of the cast: one skin tone, one sleeve, short nails, always the same side of the product. A second hand is a second card.

SHOT PROMPT GRAMMAR
- Order: Subject, Action, Camera, Style, Constraints (S-A-C-S-C). Subject first so the model binds the identity before it reads motion.
- Actions stay slow, soft and continuous. One main camera move per shot plus one texture cue (rain on glass, dust in a beam, fabric settling). Two moves in one prompt cancel each other.
- 50-80 words, three sentences, one visual promise. "Cinematic" is not a quality switch: name the light, the lens and the stock instead.
- Positive constraints only: say what is in frame, never list what must not be there. Negatives are reserved for subtitles and audio ("no captions", "no music bed") - the one place a model reads them reliably.
- Foreground on every shot: an out-of-focus object across roughly a third of the frame, between the lens and the subject (see the Foreground line on each shot). Say "blurred" or the model renders it sharp. Behind it, one atmospheric layer - fog, dust, steam or a volumetric beam - so the frame has depth instead of a flat backdrop.
- Length matches the file. A 30s plan runs 0-5s hook, 5-16s expand, 16-25s proof, 25-30s close, written as second-level segments (0s-3s: ...). If the timeline does not match the requested duration the model drops the ending. On a model that ignores seconds, fall back to shot numbers; on 2.5-class models seconds are first-class.
- Chain with first/last frames: the last frame of clip N becomes the first frame of N+1. Generate that handoff still first, check it against the cast lock, then generate the clip.
- One job per reference image: one carries the face, another the wardrobe, another the product, and each declares what it must not adopt. Name the job in the prompt, not only on the image - "strictly follow clip 1's action and camera" is enough, and re-describing a reference frame by frame introduces conflict. Stay inside the model cap (Seedance 2.5: 30 images, 10 videos, 10 audio).
- Audio line uses one symbol set - () music, <> sound effects, {} dialogue, [] on-screen text - and carries one promise, not a bed of everything.
- Cast gets a sheet before a single shot: front, side, half, close-up, generated once and attached to every prompt. Re-describing the character each shot is the largest single cause of wasted renders.
- Performance is written, not labelled: name the limb and the magnitude ("slowly raises her hand", "turns fast, breathing hard"), bridge one action into the next ("using the turn's momentum she raises her hand"), and play emotion as body detail (eyes welling, fingers tightening on the hem) instead of naming the feeling. Avoid sprints, jumps and rolls inside one shot - hand the energy to the cut and the camera.
- Light gets its own sentence: direction, quality (hard or soft), contrast. Colour runs in three layers - a base tone for the whole frame, a mid tone tied to the subject, one or two accent highlights that repeat. Then delete the word cinematic.
- Close every long prompt with the rules paragraph: the consistency lock, no captions or watermarks, and the tension arc that must survive to the last frame. It is the red line the model does not cross.
- Debug at draft resolution. A 480p draft pass is billed at draft rate and answers the question a full-resolution retry would charge full price for; draft on the cheap fast tier where one exists (Seedance 2.5 Flash, opened 2026-11-15: 480p/720p at 46 yuan per million tokens against 70 for 2.5, about twice as fast), and upgrade only the winner. Retakes eat most of a budget - roughly seven of ten credits die on a bad take - so previs the complex shot as a white model first (7-10 retries down to 2-3), shoot 3-5s pieces and stitch, and change one variable per retry.
- Gate the first frame before you scale: wrong face or product, garbled text, malformed hands, CGI look. A failed render gets diagnosed in concrete words first - material, light path, aerial perspective, overlapping structure, a negative that did not hold - then repaired with those words. A blind re-roll with a new seed doubles the bill; only after the repair fails do you change the seed.
- Read the reference post before you write: pull its transcript and overview frame sheet locally so line timing and cut rhythm come from the real post, not a guess.

SHOT BOARD
1. [0.0-2.5] macro, slow push, Hero: Extreme close-up of the real product. One edge, one material, one highlight.
   Prompt: Macro of [PRODUCT] from the real capture, extracted design tokens, one key from camera left, blurred foreground across a third of the frame, vertical 9:16, no text
   Foreground: the soft edge of a shoulder across the right third, out of focus, between the lens and the subject
   Audio: Riser into impact.
2. [2.5-6.0] 50mm, locked, Whole object: The full product, clean and exact, at correct proportion.
   Prompt: Front three-quarter of [PRODUCT], fixed proportions from the capture, one soft box, 50mm, locked, [SURFACE], vertical 9:16, no text
   Foreground: the soft edge of a shoulder across the right third, out of focus, between the lens and the subject
   Audio: Settled impact tail.
   Line: [name the product once]
3. [6.0-10.0] 35mm, lateral track, Feature one: One real feature, one motion. The camera moves, the product stays still.
   Prompt: Feature detail from the capture sheet, real design tokens, 35mm, lateral track, [LIGHT], vertical 9:16, no text
   Foreground: a hanging fabric edge, blurred across the top third, out of focus, between the lens and the subject
   Audio: UI or material sound.
   On screen: [feature name]
   Line: [the feature as a benefit, one line]
4. [10.0-14.0] 50mm, slow push, Feature two: A second feature proves the first claim. Same light logic, new angle.
   Prompt: Second feature detail, same light logic, 50mm, slow push, [SURFACE], vertical 9:16, no text
   Foreground: a blurred strand of hair across the left third, out of focus, between the lens and the subject
   Audio: Second sound, same family.
   On screen: [feature two]
   Line: [the proof in one line]
5. [14.0-18.0] 85mm, locked, Material: Finish and material under a moving light. The texture does the talking.
   Prompt: Material close of [PRODUCT], light sweep, one texture cue, 85mm, locked, vertical 9:16, no text
   Foreground: a soft foreground shape thrown out of focus across the bottom third, out of focus, between the lens and the subject
   Audio: Material sound, close.
6. [18.0-22.0] 35mm, orbit, In use: The product in a real hand, real scale. The hand enters from frame left.
   Prompt: In-use shot of [PRODUCT], real hand for scale, practical light, 35mm, orbit, vertical 9:16, no text
   Foreground: a blurred strand of hair across the left third, out of focus, between the lens and the subject
   Audio: Real use sound.
   Line: [what changes for the user]
7. [22.0-26.5] 24mm, pull back, Comparison: The product beside the thing it replaces. Both real, same light.
   Prompt: Two-shot of [PRODUCT] and [OLD OBJECT], fixed light, 24mm, pull back, [SURFACE], vertical 9:16, no text
   Foreground: the soft edge of a shoulder across the right third, out of focus, between the lens and the subject
   Audio: Comparison hit.
   On screen: [the difference in 4 words]
   Line: [the comparison, no exaggeration]
8. [26.5-30.0] 50mm, locked, Front freeze: Clean front, name and CTA typeset in the edit. The product holds still for one second.
   Prompt: Front freeze of [PRODUCT], clean [SURFACE], fixed key, 50mm, locked, vertical 9:16, text added in the edit
   Foreground: the soft edge of a shoulder across the right third, out of focus, between the lens and the subject
   Audio: Resolve chord.
   On screen: [name, price if real, CTA]
9. [30.0-33.0] 50mm, locked, End card: The same front frame with the offer line and the URL, typeset.
   Prompt: Same front freeze, clean background, space in the lower third, 50mm, locked, vertical 9:16
   Foreground: a passing guard rail, smeared by motion across the right third, out of focus, between the lens and the subject
   Audio: Tail, no new element.
   On screen: [offer and URL]
10. [33.0-35.0] macro, slow pull, Loop: A macro detail that rhymes with shot one, so the film loops cleanly.
   Prompt: Macro detail of [PRODUCT], same key as shot 01, 35mm, slow pull, vertical 9:16, no text
   Foreground: the soft edge of a shoulder across the right third, out of focus, between the lens and the subject
   Audio: Riser tail.

MOODBOARD PROMPTS
- Styleframe (16:9): Two or three stills locking palette, type, light and camera before any motion. The chosen frame becomes the reference for every shot.
- Motion sheet (16:9): Energy and tone axes mapped from the real brand, with one motion rule stated in words.
- Material board (1:1): Three material swatches from the product capture: finish, texture, edge. Used to keep the model honest.
- Type lock (9:16): The type system on a clean frame: scale, tracking, safe area, one accent. Typeset, never generated.
- QA frame pair (16:9): Two fixed acceptance frames per shot: one at the start of the move, one at the end, checked against the product capture.
- Token sheet (16:9): Three swatches with hex values, two type styles with weights, one motion rule in a sentence. This is the contract the renderer reads, not a mood.
- Silhouette check (1:1): The product reduced to a black silhouette at the correct proportion, beside the real capture silhouette. If they do not overlay, the render is wrong before it starts.
- Light study (16:9): The same product under the key, the fill and the edge, one frame each, so the chosen setup is a decision rather than a prompt adjective.
- Cut map (16:9): Every shot on one strip with its beat, its length and its motion rule. The paid cut is marked on the same strip, not built separately.
- Feature map (16:9): Every feature on the product, numbered, with the shot that carries it beside it. A feature without a shot is a feature the film does not claim.
- Loop frame (9:16): The last frame beside the first, checked so the film loops without a visible seam on a feed.
- Hand scale (1:1): The product beside a real hand and a coin, shot at the same distance. Scale is checked against two familiar objects, not one.
- Ratio pair (16:9): The same shot framed for 16:9 and 9:16 side by side, with the product's safe area marked. The vertical frame is composed, not cropped.

SOUND AND MUSIC
Design the film as one arc: a riser into the hero, a clean impact, then material and UI sounds that belong to the product. Music is minimal and low; the product sounds carry the proof. No stock whoosh on a cut that does not need one. Loudness -16 LUFS for web, -14 for paid social, true peak -1 dB. Build the sound from the product: the lid, the clasp, the pour, the click. Those recordings are the proof the object is real. Music is one bed with a single lift at the feature that matters; if the bed is doing the persuading, the shot is not.

QC GATE
- Proportion: Silhouette and scale match the product capture; on fail: A stretched or resized product: discard the take
- Label: Label, logo and finish are the real ones; on fail: Redrawn text: rebuild from the capture
- Colour: Product colour matches the swatch under the stated light; on fail: A hue shift: re-render with the token
- Text: All text is typeset with real fonts and reads at 390 px; on fail: Model-drawn glyphs: delete the layer
- Motion: One motion rule holds for the whole film; on fail: Camera and product moving together: re-board
- Material: Texture reads as the real material, not a generic glossy surface; on fail: Plastic sheen: change the light, not the prompt adjective
- Seams: Cuts land on the beat or on a motion match; on fail: A visible jump: trim and re-join
- Loudness: Master hits the target with no clipping; on fail: Over: re-mix
- Delivery: Both ratios, captions where spoken, AI label where synthetic; on fail: Missing asset: block delivery
- Silhouette: The rendered silhouette overlays the capture silhouette; on fail: A mismatch: fix proportion before lighting
- Font: Every typeface is licensed for commercial use and recorded; on fail: An unlicensed face: replace it before delivery
- Cut: One transition type, cuts on beat or motion match; on fail: A stray transition: remove it
- Surface: The surface material matches the stated finish in the prompt; on fail: A generic glossy floor: change the surface, not the adjective

FAILURE ATLAS
- Product looks like a different object (Prompt described a category instead of the capture) -> Paste the capture and the proportion lock, regenerate
- Text is garbled (The model was asked to draw type) -> Remove text from the prompt, typeset in the edit
- Film feels like a slideshow (Every shot is locked and the same distance) -> Add one move per feature and vary the lens
- Colour drifts between shots (The palette was paraphrased) -> Use the token names verbatim and re-render
- Product floats (No surface or hand for scale) -> Add the surface lock and a real hand
- Cuts feel random (No beat map) -> Mark the music beats before the edit and cut on them
- Paid cut loses the product (The 9:16 was cropped from 16:9) -> Re-frame the board for 9:16 and re-render the used shots
- Product reads as a render (Scale was described in adjectives instead of a ratio) -> Add the ratio lock and a real hand for reference
- Film feels cheap (Too many transitions and no sound from the object) -> Cut with the cut, and record the product's own sounds
- Paid cut loses the label (The 9:16 was cropped from the centre of 16:9) -> Re-frame the board for 9:16 and re-render the used shots
- Feature shots look alike (Every feature was shot at the same distance) -> Vary lens and distance per feature, and give each one its own motion

STEPS
1. Capture the product page: images, feature list, materials, dimensions, label artwork. Record the URL and the date.
2. Extract three colour swatches, two type styles and one motion rule. Write them as tokens, not adjectives.
3. Make two or three styleframes and pick one. Nothing moves until the styleframe is approved.
4. Map each feature to one shot and write the board: lens, move, action, on-screen text and line.
5. Price the board and set a cap. Cut shots before cutting quality if the quote is over.
6. Render the first shot from a fixed seed and compare it to the capture. Only then render the rest.
7. Check two acceptance frames per shot: start of move and end of move.
8. Assemble with temp music, then replace the music and cut on the beats.
9. Typeset every word in the edit with the locked fonts and safe area.
10. Re-render only failed shots, at their used lengths. Never re-render the whole film to fix one frame.
11. Export 16:9 and 9:16 from the same board, with captions and the AI label.
12. Write the delivery manifest: ratios, codecs, loudness, fonts, seeds and the accepted takes.
13. Write the token sheet as values: hex colours, font names and weights, one motion rule. Nothing in the film may use a value that is not on the sheet.
14. Overlay the rendered silhouette on the capture silhouette before lighting a single frame.
15. Record the product's own sounds before the edit and build the mix from them.
16. Mark the paid cut on the same strip as the master and re-frame, never crop.
17. Number every feature and name the shot that carries it before the board is priced.
18. Check the loop frame against the first frame before the final export.

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 - Product Film
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.
- 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.
- 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.
- 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.

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 (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.
- 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.
- 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.
- 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 CONTRACT
- Product understanding brief
- Styleframes locking the look
- Feature-to-shot map
- Deterministic shot renders
- Two QA frames per shot
- Independent final review and delivery

CHECKS
- Tokens come from the real product page
- The styleframe is approved before any motion
- One feature per shot, one motion per shot
- Every text element is typeset, never generated
- Two acceptance frames per shot
- Seeds and model versions are recorded per shot
- Both ratios are re-framed, not cropped
- Loudness and true peak hit the target
- The delivery manifest is complete
- Every value in the film appears on the token sheet
- The silhouette overlays the capture before any lighting work
- The mix is built from recorded product sounds
- Every typeface has a recorded licence
- Every claimed feature has a shot
- The film loops without a visible seam

REFUSE OR STOP WHEN
- No invented brand, feature or price
- No model-drawn text
- No shot before the styleframe is approved
- No re-render of a passing shot
- No delivery without the manifest and the AI label

Render budget, today

35s on the board · Seedance 2.5 720p

$6.47cheapest seller $8.71median $16.55dearest

10 shots, 35 seconds of finished screen time, priced at today's published Seedance 2.5 720p rates. That is the render bill only: no retries, stills, music, voice or edit. The rate board carries every seller, the date each price was read and the calculator for a different length or resolution.

The stack

what each role uses

RoleToolLink
Script and shot logicClaudefree
KeyframesGPT Imagefree
Camera and plateHiggsfieldvendor
MotionRunwayvendor
UpscaleMagnificvendor
FinishTopazvendor
EditCapCutfree