Studio · UGC · 9:16 · 100 ads a day
UGC Ad Factory
A repeatable UGC ad line: research the winning angle, write for one sub-avatar, lock the actor, then batch hooks while the body stays fixed. One shoot day, a hundred ads, every one written for a different buyer and every claim traceable to the product page or the operator's own record.
The whole line, one pack
story to stitch
The board below is the core, not a ceiling; extend it until the story lands.
What you get
11 deliverables
- Angle brief: proven angle, sub-avatar, private fear and the one promise
- Script in hook, problem, mechanism, proof, objection, offer, CTA
- Actor lock with recorded consent and a product multiview
- One approved master ad before any batch is generated
- Hook bank: 20 openers written against the same fixed body
- Word-level captions and a platform-safe text layer
- Compliance pass: every claim mapped to its evidence
- Rank sheet: hook rate, hold rate, click and cost per result per variant
- Hook bank of 20 openers, each written for the same sub-avatar and the same body
- Claim map: every spoken claim with its evidence and its on-screen source
- Batch sheet: variant id, hook, spend, result and verdict per ad
World and look
paste before anything
| World | YOUR WORLD - the buyer's real room, not a studio. Kitchen counter, bathroom shelf, car seat, gym bag. One location per ad, stated once and held for every shot in that ad. |
| Palette | YOUR PALETTE - phone-camera naturalism: available light, real skin texture, one practical source. The product label must stay readable at thumbnail size. No cinematic grade, no teal-orange. |
| Negative | extra fingers, fused hands, face morph, changing eye colour, plastic skin, warped jewellery, floating earrings, captions, watermark, logo, extra teeth |
Locks
location, props, wardrobe, axis
| actor | One actor card: age, face, hair, one distinguishing mark, one garment, one voice. Consent recorded before the first frame. The same card is pasted into every prompt in the batch. |
| product | Product multiview: front, back, label, cap, size in hand. The label is never redrawn by the model; if it drifts, the take is dead. |
| location | One room per ad, with the light source named. A second room is a second ad, not a second setup inside the same ad. |
| wardrobe | One garment per batch so variants read as the same person. A change of garment is a new actor card. |
| claims | Every claim carries its evidence: the product page line, the lab result, the operator's own measurement. No evidence, no claim. |
| light | One practical source per ad, named in the prompt: window left, window right, or the bathroom bulb. A second source is a second look and a second ad. |
| frame | Phone at chest height, arm's length, subject centred with headroom for the platform UI. The same frame across a batch so variants differ only in the hook. |
| props | The supporting objects the ad touches: the towel, the receipt, the shelf. Each is named in the script and stays in the same place between shots, because a moved object reads as a different room. |
Every lock is pasted into the prompt for every shot that touches it. A lock that is paraphrased is not a lock.
Cast locks
1 identities
ACTOR
Fill from your own vault card: fixed age, face, hair, one distinguishing mark.
One signature garment held identical across every shot and episode.
Fixed voice note for this role: pace, tone, one verbal habit.
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
6 skills, in order
| Skill | Job in this run |
|---|---|
avatar-research | brand voice, avatar and subavatar |
customer-research | customer and market evidence |
ad-copy | platform-native ad scripts |
ad-creative | ad copy at scale |
video | shots, motion, editing and the cut |
compliance-checker | pre-spend policy gate |
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
| Source | What it gives |
|---|---|
| TrendTrack | winning ads, hooks, spend and engagement, creative library |
| GetHookd | winning ads, hooks, scripts and transcripts |
| Foreplay | swipe file, ad copy, video downloads and timestamps |
| Meta Ad Library | live ads, copy, start dates and platforms |
| TikTok Creative Center | top ads, hooks, music and formats |
| Paste, CSV or screenshot | any ad the buyer can see and has the right to study |
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
8 core shots · full prompts
0.0-3.0s
24mm
handheld push
Hook — [ACTOR] addresses camera mid-thought, as if the viewer interrupted. The product is not in frame yet.
[ACTOR LOCK] selfie framing at arm's length, [LOCATION CARD], natural window light, 24mm, handheld push, real skin texture, vertical 9:16, no text
foregroundthe soft edge of a shoulder across the right third, out of focus, between the lens and the subject
audioDirect voice, room tone, no music.
on screen[hook line, 6 words max]
line[hook line, spoken]
3.0-8.0s
35mm
locked
Problem — [ACTOR] shows the problem in the room: the shelf, the drawer, the mirror. The pain is visible, not narrated.
[ACTOR LOCK] in [LOCATION CARD] showing [PROBLEM OBJECT], 35mm, locked, practical light, vertical 9:16
foregrounda blurred A-pillar across the left third, out of focus, between the lens and the subject
audioVoice continues, one real sound of the problem.
line[the pain in the buyer's words]
8.0-14.0s
50mm
slow push
Mechanism — A visible cause and effect with [PRODUCT CARD]: the thing it does, in one motion.
[PRODUCT CARD] in use by [ACTOR LOCK], [LOCATION CARD], 50mm, slow push, label readable, vertical 9:16
foregrounda blurred A-pillar across the left third, out of focus, between the lens and the subject
audioProduct sound close, voice under it.
line[the mechanism in one sentence]
14.0-20.0s
50mm
locked
Proof — One substantiated detail, held long enough to be checked: texture, seam, fit, reading on a screen.
Macro of [PRODUCT CARD] detail, [ACTOR LOCK] hand for scale, 50mm, locked, one key light, vertical 9:16
foregrounda blurred strand of hair across the left third, out of focus, between the lens and the subject
audioVoice drops, one detail sound.
on screen[evidence line with source]
line[the proof, with its source named]
20.0-25.0s
35mm
handheld
Objection — [ACTOR] answers the comment-section objection directly to camera, no defensiveness.
[ACTOR LOCK] to camera, [LOCATION CARD], 35mm, handheld, natural light, vertical 9:16
foregroundthe soft edge of a shoulder across the right third, out of focus, between the lens and the subject
audioDirect voice, no music.
on screen[objection, quoted]
line[the answer in one sentence]
25.0-28.0s
24mm
locked
Offer — Product on the counter with the real price card. Terms stated once, no countdown.
[PRODUCT CARD] on [SURFACE], price card in frame, 24mm, locked, practical light, vertical 9:16
foregroundthe soft edge of a table corner across the right third, out of focus, between the lens and the subject
audioVoice states terms, one button.
on screen[real price and terms]
line[the offer, exactly as sold]
28.0-30.0s
24mm
locked
CTA — [ACTOR] holds the product at chest height, one gesture to the lower third.
[ACTOR LOCK] holding [PRODUCT CARD], chest height, 24mm, locked, [LOCATION CARD], vertical 9:16
foregroundthe out-of-focus dash edge across the bottom third, out of focus, between the lens and the subject
audioVoice CTA, music button.
on screen[CTA, 4 words max]
line[one light CTA]
30.0-32.0s
24mm
locked
Loop — A frame that rhymes with the hook so the replay is seamless.
[ACTOR LOCK] same framing as shot 01, product now visible, 24mm, locked, vertical 9:16
foregrounda hanging fabric edge, blurred across the top third, out of focus, between the lens and the subject
audioMusic tail, no voice.
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
| Time | Beat | Job |
|---|---|---|
| 0-3s | Hook | One line for one sub-avatar, spoken before the brain can scroll. Name the private fear, not the category. |
| 3-8s | Problem | The pain in the buyer's own words. Show the moment, do not describe the market. |
| 8-14s | Mechanism | A visible cause and effect with the product. One mechanism, shown once. |
| 14-20s | Proof | One substantiated detail, shown not claimed: texture, fit, before-after, third-party result. |
| 20-25s | Objection | Answer the objection the comments actually repeat, in one sentence. |
| 25-28s | Offer | Real price and terms from the operator only. Never invent a discount or a deadline. |
| 28-30s | CTA | One light call to action at the offer moment, on screen and spoken once. |
| 30-32s | Loop | The last frame rhymes with the first so a replay feels intentional. |
Moodboard prompts
11 images
Hooks and variants
10 openers
- Hook on the objection instead of the pain: start with what the buyer does not believe.
- Hook on the proof: open on the close-up result, then explain how.
- Hook on the price: name the real number in the first two seconds and hold the rest.
- Swap the sub-avatar: same body, new private fear, new hook only.
- Silent cut: captions carry the whole script for autoplay.
- Founder cut: the operator speaks the same script, no actor.
- Comment-reply cut: the ad opens as a reply to a real comment, then answers it.
- Two-actor split: the same script delivered by a second consented actor, tested against the first.
- Receipt cut: the ad opens on the real order confirmation and explains the product from there.
- Stitch cut: the ad opens on a customer clip and answers it with the product.
Sound and music
direction, not a playlist
Voice first: recorded on the phone or generated, always dry and close. Music is a bed, ducked 12 dB under the voice, and it never carries the hook. Real product sounds (click, pour, zip) sit above the bed and below the voice. Keep the voice stem separate so a platform cut can re-mix without a re-record. Record the voice on the phone when the actor is real and keep that file: it is the reference for every generated line. Bed music sits 12 dB under the voice and never enters the first two seconds. Product sounds are captured rather than synthesised, because the click is the proof.
QC gate
14 checks, pass or fail
| Check | Pass | Fail |
|---|---|---|
| Claim evidence | Every claim maps to a product-page line, a lab result or the operator's measurement | An unsupported claim: cut it or mark it as opinion |
| Hook clarity | The first line names the sub-avatar's fear in under 3 seconds | A vague hook: rewrite, do not re-render |
| Label | Product label matches the multiview in every frame | Drift: discard the take |
| Actor | Face, hair and garment match the actor card | Drift: regenerate from the card |
| Caption sync | Word-level captions match the spoken line | Drift: rebuild from the script |
| Caption legibility | Text reads at 390 px and clears the platform UI | Overlap: move the text block |
| Audio | Voice intelligible, music ducked, no clipping | A clip: re-mix before export |
| Terms | Price and terms match the operator's record exactly | A mismatch: block delivery |
| Policy | No prohibited claim, no before-after that implies a medical result, no fake urgency | A violation: rewrite and re-shoot the line |
| Disclosure | AI label where synthetic, paid-partnership label where applicable | Missing: block delivery |
| Consent | Actor consent is on file and the card matches the person on screen | No consent: stop the batch |
| Batch delta | Each variant differs only in the declared slot | A second change: discard the variant |
| Result | Every variant has a recorded result, not an impression | No result: the variant does not count |
| Hook frame | The first frame reads at thumbnail size with no sound | An unreadable frame: re-shoot the hook, not the body |
What would make this fail
12 known failures
| Symptom | Cause | Fix |
|---|---|---|
| Hook rate is flat | The hook names a category, not a private fear | Rewrite 20 hooks against one sub-avatar and test again |
| Ad feels like an ad | Script uses marketer language instead of the buyer's words | Mine the comments and reviews, rewrite in their phrasing |
| Product looks fake | Label was generated instead of captured | Use the multiview capture and composite the real label |
| Comments attack the claim | Proof was asserted, not shown | Replace the claim with one substantiated detail |
| Cost per result climbs | Too many variants, too few hooks | Freeze the body, batch hooks only |
| Actor changes between variants | New seed or a paraphrased card | Paste the card and reuse the accepted seed |
| Platform rejects the ad | A prohibited claim or a fake countdown | Rewrite the line and re-run the compliance pass |
| Batch sounds identical | Same read, same pace, same music | Vary the first three seconds only; keep the body fixed |
| Every variant performs the same | The hooks were paraphrases of one idea | Write hooks from six different objections, not six synonyms |
| Batch cannot be compared | Two things changed between variants | Freeze the body and re-render with one slot changed |
| Account gets flagged | A claim or a countdown the operator could not support | Cut the line and re-run the compliance pass before republishing |
| Ad is skipped at two seconds | The hook line arrives after the face has already been read | Cut the first frame to the hook, not to a greeting |
Agent runbook
15 steps
Produce UGC ads as a line, not a one-off: research the proven angle, name one sub-avatar and one private fear, lock the actor and the product, approve one master, then batch hooks while the body stays fixed. Every claim is traceable and every variant is measured.
Collect the evidence: product page, real reviews, comment threads, the operator's own numbers. Record the URL and the date.
Name one sub-avatar and the private fear. Write the angle in one sentence and the promise in one sentence.
Write the script in hook, problem, mechanism, proof, objection, offer, CTA. Mark every claim with its evidence.
Lock the actor with recorded consent and build the product multiview. Approve both before generation.
Generate one master ad. Check hook, label, captions, audio and claims against the QC table.
Only after the master passes, batch 20 hooks against the same body. Change nothing else.
Run the compliance pass on every variant: claims, urgency, before-after, disclosure.
Publish in small batches and record hook rate, hold rate, click and cost per result per variant.
Kill the bottom half on measured results, keep the top three and write the next 20 hooks from what won.
Archive the winning script, card, seed and evidence in the project file before the next batch.
Write the hook bank before the shoot: twenty openers, each naming a different private fear of the same sub-avatar.
Shoot or generate the body once and freeze it. Every variant is a hook patch on that body, never a re-shoot.
Record spend and result per variant in the batch sheet the day it runs, while the number is still clean.
Write the next twenty hooks from the three that won, and retire the angle that lost twice.
Check the second frame before the rest of the ad: it must give a reason to keep watching.
The prompt
paste into Claude or Codex
You are the production agent for the AISLE playbook "UGC Ad Factory".
MISSION
Produce UGC ads as a line, not a one-off: research the proven angle, name one sub-avatar and one private fear, lock the actor and the product, approve one master, then batch hooks while the body stays fixed. Every claim is traceable and every variant is measured.
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. Buyer and angle: Name the buyer, the objection and the proof before a single frame. One ad, one buyer.
02. Hook bank: Write ten hooks per angle; keep the three that survive a three-second read.
03. Creator lock: One original AI-generated creator per ad: face sheet, wardrobe, voice, behaviour. No stock face, no borrowed likeness.
04. Board: Every ad gets a hook, timecode, lens, action, claim and prompt. Direct the scroll-stop, not the vibe.
05. Generate: Keyframes, image-to-video, dialogue, captions and SFX. Three hooks; keep the one that holds three seconds.
06. Consistency: Same creator, product, wardrobe and light within an ad. Vary the buyer, never the truth.
07. Compliance: No fake testimonial, no fake receipt, no cure claim, no before/after that implies a medical outcome.
08. Cut: Timeline, captions, sound, delivery in 9:16, 16:9 and 1:1. The ad is the product.
09. Batch: Name every ad by buyer, angle and hook. Ship the batch, not the hero.
Receipt for every ad: model, seconds, cost, hook. A cap per batch that is enforced. No claim ships without evidence.
WORLD
YOUR WORLD - the buyer's real room, not a studio. Kitchen counter, bathroom shelf, car seat, gym bag. One location per ad, stated once and held for every shot in that ad.
Palette: YOUR PALETTE - phone-camera naturalism: available light, real skin texture, one practical source. The product label must stay readable at thumbnail size. No cinematic grade, no teal-orange.
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)
- ACTOR (Identity slot): ACTOR: [fixed age], [fixed face], [fixed hair], [signature garment] - consent recorded, locked across the batch.
LOCKS (location, props, wardrobe, axis)
- actor: One actor card: age, face, hair, one distinguishing mark, one garment, one voice. Consent recorded before the first frame. The same card is pasted into every prompt in the batch.
- product: Product multiview: front, back, label, cap, size in hand. The label is never redrawn by the model; if it drifts, the take is dead.
- location: One room per ad, with the light source named. A second room is a second ad, not a second setup inside the same ad.
- wardrobe: One garment per batch so variants read as the same person. A change of garment is a new actor card.
- claims: Every claim carries its evidence: the product page line, the lab result, the operator's own measurement. No evidence, no claim.
- light: One practical source per ad, named in the prompt: window left, window right, or the bathroom bulb. A second source is a second look and a second ad.
- frame: Phone at chest height, arm's length, subject centred with headroom for the platform UI. The same frame across a batch so variants differ only in the hook.
- props: The supporting objects the ad touches: the towel, the receipt, the shelf. Each is named in the script and stays in the same place between shots, because a moved object reads as a different room.
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. [0.0-3.0] 24mm, handheld push, Hook: [ACTOR] addresses camera mid-thought, as if the viewer interrupted. The product is not in frame yet.
Prompt: [ACTOR LOCK] selfie framing at arm's length, [LOCATION CARD], natural window light, 24mm, handheld push, real skin texture, 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: Direct voice, room tone, no music.
On screen: [hook line, 6 words max]
Line: [hook line, spoken]
2. [3.0-8.0] 35mm, locked, Problem: [ACTOR] shows the problem in the room: the shelf, the drawer, the mirror. The pain is visible, not narrated.
Prompt: [ACTOR LOCK] in [LOCATION CARD] showing [PROBLEM OBJECT], 35mm, locked, practical light, vertical 9:16
Foreground: a blurred A-pillar across the left third, out of focus, between the lens and the subject
Audio: Voice continues, one real sound of the problem.
Line: [the pain in the buyer's words]
3. [8.0-14.0] 50mm, slow push, Mechanism: A visible cause and effect with [PRODUCT CARD]: the thing it does, in one motion.
Prompt: [PRODUCT CARD] in use by [ACTOR LOCK], [LOCATION CARD], 50mm, slow push, label readable, vertical 9:16
Foreground: a blurred A-pillar across the left third, out of focus, between the lens and the subject
Audio: Product sound close, voice under it.
Line: [the mechanism in one sentence]
4. [14.0-20.0] 50mm, locked, Proof: One substantiated detail, held long enough to be checked: texture, seam, fit, reading on a screen.
Prompt: Macro of [PRODUCT CARD] detail, [ACTOR LOCK] hand for scale, 50mm, locked, one key light, vertical 9:16
Foreground: a blurred strand of hair across the left third, out of focus, between the lens and the subject
Audio: Voice drops, one detail sound.
On screen: [evidence line with source]
Line: [the proof, with its source named]
5. [20.0-25.0] 35mm, handheld, Objection: [ACTOR] answers the comment-section objection directly to camera, no defensiveness.
Prompt: [ACTOR LOCK] to camera, [LOCATION CARD], 35mm, handheld, natural light, vertical 9:16
Foreground: the soft edge of a shoulder across the right third, out of focus, between the lens and the subject
Audio: Direct voice, no music.
On screen: [objection, quoted]
Line: [the answer in one sentence]
6. [25.0-28.0] 24mm, locked, Offer: Product on the counter with the real price card. Terms stated once, no countdown.
Prompt: [PRODUCT CARD] on [SURFACE], price card in frame, 24mm, locked, practical light, vertical 9:16
Foreground: the soft edge of a table corner across the right third, out of focus, between the lens and the subject
Audio: Voice states terms, one button.
On screen: [real price and terms]
Line: [the offer, exactly as sold]
7. [28.0-30.0] 24mm, locked, CTA: [ACTOR] holds the product at chest height, one gesture to the lower third.
Prompt: [ACTOR LOCK] holding [PRODUCT CARD], chest height, 24mm, locked, [LOCATION CARD], vertical 9:16
Foreground: the out-of-focus dash edge across the bottom third, out of focus, between the lens and the subject
Audio: Voice CTA, music button.
On screen: [CTA, 4 words max]
Line: [one light CTA]
8. [30.0-32.0] 24mm, locked, Loop: A frame that rhymes with the hook so the replay is seamless.
Prompt: [ACTOR LOCK] same framing as shot 01, product now visible, 24mm, locked, vertical 9:16
Foreground: a hanging fabric edge, blurred across the top third, out of focus, between the lens and the subject
Audio: Music tail, no voice.
MOODBOARD PROMPTS
- Actor lock (3:4): Four views of the actor in the same light and garment: face, three-quarter, full body, hands with product. Consent recorded before this sheet exists.
- Product multiview (1:1): Front, back, label, cap and size-in-hand on a neutral surface. The label is the reference; nothing may be redrawn.
- Room plate (9:16): The buyer's real room with the light source named. No studio backdrop, no styled shelf.
- Hook frame (9:16): The first frame of the ad: face, product absent, one line of typeset text in the lower third.
- Caption kit (9:16): Two caption styles: one word-by-word pop and one clean lower third. Same font, same weight, checked at 390 px.
- Hook frames (9:16): Six first frames side by side, same actor, same room, six different hooks. Judge them at thumbnail size, because that is where the scroll decision happens.
- Objection wall (16:9): The real comments that repeat, quoted verbatim on one board. This is the script for the objection beat, not a guess about what buyers think.
- Claim map (16:9): Every claim in the script on one line with its evidence beside it: page line, lab result or the operator's own measurement. A blank evidence column means the line gets cut.
- Batch sheet (16:9): One row per variant: hook, first frame, spend, hook rate, cost per result, verdict. This sheet decides the next twenty hooks.
- Second frame (9:16): The frame after the hook, checked for a reason to keep watching: a hand moving, a product entering, a face changing. A static second frame loses the viewer the hook earned.
- Price card (9:16): The real offer at final size with the terms beside it. If the price needs a smaller font to be acceptable, the offer is the problem.
SOUND AND MUSIC
Voice first: recorded on the phone or generated, always dry and close. Music is a bed, ducked 12 dB under the voice, and it never carries the hook. Real product sounds (click, pour, zip) sit above the bed and below the voice. Keep the voice stem separate so a platform cut can re-mix without a re-record. Record the voice on the phone when the actor is real and keep that file: it is the reference for every generated line. Bed music sits 12 dB under the voice and never enters the first two seconds. Product sounds are captured rather than synthesised, because the click is the proof.
QC GATE
- Claim evidence: Every claim maps to a product-page line, a lab result or the operator's measurement; on fail: An unsupported claim: cut it or mark it as opinion
- Hook clarity: The first line names the sub-avatar's fear in under 3 seconds; on fail: A vague hook: rewrite, do not re-render
- Label: Product label matches the multiview in every frame; on fail: Drift: discard the take
- Actor: Face, hair and garment match the actor card; on fail: Drift: regenerate from the card
- Caption sync: Word-level captions match the spoken line; on fail: Drift: rebuild from the script
- Caption legibility: Text reads at 390 px and clears the platform UI; on fail: Overlap: move the text block
- Audio: Voice intelligible, music ducked, no clipping; on fail: A clip: re-mix before export
- Terms: Price and terms match the operator's record exactly; on fail: A mismatch: block delivery
- Policy: No prohibited claim, no before-after that implies a medical result, no fake urgency; on fail: A violation: rewrite and re-shoot the line
- Disclosure: AI label where synthetic, paid-partnership label where applicable; on fail: Missing: block delivery
- Consent: Actor consent is on file and the card matches the person on screen; on fail: No consent: stop the batch
- Batch delta: Each variant differs only in the declared slot; on fail: A second change: discard the variant
- Result: Every variant has a recorded result, not an impression; on fail: No result: the variant does not count
- Hook frame: The first frame reads at thumbnail size with no sound; on fail: An unreadable frame: re-shoot the hook, not the body
FAILURE ATLAS
- Hook rate is flat (The hook names a category, not a private fear) -> Rewrite 20 hooks against one sub-avatar and test again
- Ad feels like an ad (Script uses marketer language instead of the buyer's words) -> Mine the comments and reviews, rewrite in their phrasing
- Product looks fake (Label was generated instead of captured) -> Use the multiview capture and composite the real label
- Comments attack the claim (Proof was asserted, not shown) -> Replace the claim with one substantiated detail
- Cost per result climbs (Too many variants, too few hooks) -> Freeze the body, batch hooks only
- Actor changes between variants (New seed or a paraphrased card) -> Paste the card and reuse the accepted seed
- Platform rejects the ad (A prohibited claim or a fake countdown) -> Rewrite the line and re-run the compliance pass
- Batch sounds identical (Same read, same pace, same music) -> Vary the first three seconds only; keep the body fixed
- Every variant performs the same (The hooks were paraphrases of one idea) -> Write hooks from six different objections, not six synonyms
- Batch cannot be compared (Two things changed between variants) -> Freeze the body and re-render with one slot changed
- Account gets flagged (A claim or a countdown the operator could not support) -> Cut the line and re-run the compliance pass before republishing
- Ad is skipped at two seconds (The hook line arrives after the face has already been read) -> Cut the first frame to the hook, not to a greeting
STEPS
1. Collect the evidence: product page, real reviews, comment threads, the operator's own numbers. Record the URL and the date.
2. Name one sub-avatar and the private fear. Write the angle in one sentence and the promise in one sentence.
3. Write the script in hook, problem, mechanism, proof, objection, offer, CTA. Mark every claim with its evidence.
4. Lock the actor with recorded consent and build the product multiview. Approve both before generation.
5. Generate one master ad. Check hook, label, captions, audio and claims against the QC table.
6. Only after the master passes, batch 20 hooks against the same body. Change nothing else.
7. Run the compliance pass on every variant: claims, urgency, before-after, disclosure.
8. Publish in small batches and record hook rate, hold rate, click and cost per result per variant.
9. Kill the bottom half on measured results, keep the top three and write the next 20 hooks from what won.
10. Archive the winning script, card, seed and evidence in the project file before the next batch.
11. Write the hook bank before the shoot: twenty openers, each naming a different private fear of the same sub-avatar.
12. Shoot or generate the body once and freeze it. Every variant is a hook patch on that body, never a re-shoot.
13. Record spend and result per variant in the batch sheet the day it runs, while the number is still clean.
14. Write the next twenty hooks from the three that won, and retire the angle that lost twice.
15. Check the second frame before the rest of the ad: it must give a reason to keep watching.
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 - UGC 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.
- avatar-research - brand voice, avatar and subavatar: Mine reviews, forums, 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.
- customer-research - customer and market evidence: Run interviews, surveys and review mining; turn raw voice-of-customer into a persona the whole shoot is aimed at.
- 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.
- ad-creative - ad copy at scale: Generate and iterate hooks, primary text and headlines into variants, each with a falsifiable test hypothesis.
- 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 (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.
- 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.
- 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
- Winning-angle research note
- Sub-avatar and private-fear brief
- Locked actor and product sheets
- Batch of hook variants with a fixed body
- Word-level captions and compliance pass
- Ranked candidates
CHECKS
- One sub-avatar and one private fear per batch
- Every claim has a named source and a date
- Actor consent is recorded before the first frame
- The product label matches the multiview in every frame
- The master is approved before any hook is batched
- Captions are word-level and checked at 390 px
- Price and terms come from the operator, never invented
- Every variant passes the compliance pass
- Results are recorded per variant, not per campaign
- Twenty hooks exist before the batch runs
- The body is identical across variants
- Spend and result are recorded per variant on the day it runs
- The losing angle is retired rather than reworded
- The hook frame is legible as a thumbnail with the sound off
REFUSE OR STOP WHEN
- No invented price, discount, deadline or testimonial
- No medical or guaranteed-result claim
- No actor likeness without recorded consent
- No generated product label
- No batch before the master passes QC
- No delivery without the disclosure line
Render budget, today
32s on the board · Seedance 2.5 720p
8 shots, 32 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
| Role | Tool |
|---|---|
| Script and matrix | Claude |
| Actor-led UGC | Arcads |
| Avatar fallback | HeyGen |
| Voice | ElevenLabs |
| Edit | CapCut |
| Captions | Submagic |
| Cutdowns | OpusClip |
Other packs
8 more
AI Drama Series
A recurring cast, an episode arc, and the same face in every shot.
10 shots / 60s episode · cast to scriptSong VSL · 9:16 · 12-stage journeyDrama Song Ad
A drama-style sung VSL, from script and song to timing map and locked cast.
8 shots / 30s song · cast to scriptCartoon · 9:16 · 3-7s fragmentsAI Cartoon Short
A cartoon with model sheets that survive every cut.
8 shots / 40s short · cast to scriptProduct · 9:16 + 16:9 · styleframe-firstProduct Film
A product hero film built from real product pages, not a storyboard guess.
8 shots / 30s film · cast to script