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

Studio · Documentary · 16:9 · narration-led

Faceless Documentary

A long-form documentary line: article in, chapters and shots planned, narration locking shot length, footage retrieved, templates decorated, per-shot QC and a single stitch. Every factual claim carries a source, every shot carries one job, and the narration is written before a single frame is generated.

10 chapters / 10 min core boardcast to script14 moodboards9 variantsNo ceilingNo ceiling

The whole line, one pack

story to stitch

01 · Question and evidenceWrite the question, the answer and the three pieces of evidence that prove it. Record every source with URL and date.
02 · Reconstruction ruleMark what is archive, what is reconstruction and what is graphic. Label every generated image as a reconstruction.
03 · LookArchive texture, reconstruction palette, map and type language. Decide what carries the argument.
04 · NarratorCast the voice, write the read and time it to the cut. No presenter on camera.
05 · BoardEvery shot gets a timecode, source, claim, narration line and prompt. The script is the spine.
06 · GenerateReconstructions, maps, motion graphics and narration. Three takes; keep the one that survives fact-check.
07 · ConsistencySame narrator, type system, grade and reconstruction rule in every scene.
08 · SoundNarration, archive ambience and one score bed, ducked under the read.
09 · Cut and deliverTimeline, captions, source list, delivery in 16:9 and 9:16. The cut is the argument.

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

What you get

11 deliverables

  • Source record: URL, read date, author and the claim list
  • Chapter plan with one question per chapter
  • Narration script with a source marker on every factual claim
  • Shot plan: one shot per narration beat, duration locked by the read
  • Footage register with licence and provenance per clip
  • Template binding per shot and a stitch map
  • Per-shot QC: text, render, claim and audio
  • Master, captions and a source list in the description
  • Claim ledger: one row per claim with sentence, number, unit, period, URL and read date
  • Licence register: source, licence type, resolution and expiry per clip
  • Read-through timing sheet: paragraph durations that set the shot lengths

World and look

paste before anything

WorldYOUR SOURCE - the real article or script with the URL and read date recorded. A blog is corroboration, not proof. The world of the film is the evidence itself: documents, maps, measurements, archive footage with a licence.
PaletteYOUR PALETTE - documentary restraint: a neutral base, one accent for the claim under examination, and a distinct treatment for archive or reconstruction. Reconstructions are labelled on screen.
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

sourceOne source record per claim: URL, date, author, the exact sentence. A claim without a record is cut, not softened.
footageEvery clip carries a licence, a source URL and a resolution. Archive and reconstruction are labelled; stock is never presented as evidence.
templateOne decoration template per chapter, bound to the shot before render. The template may change between chapters, never inside one.
narrationOne voice, one pace, one pronunciation guide for names. The read locks the shot duration; the picture never cuts the sentence.
claimsNumbers are quoted with their unit, period and source. A derived number shows its arithmetic on screen.
readingNumbers are read the way they are written: unit, period, then the figure. A narrator who says up three hundred percent when the graphic says 3x has created a contradiction the audience will catch.
musicOne bed, one motif per chapter, no sting on a number. The bed drops under every claim and returns in the gaps, so the ear knows when to listen.

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
customer-researchcustomer and market evidence
copywritinghooks, headlines and page copy
videoshots, motion, editing and the cut
socialorganic and platform format
copy-editingcopy and caption revision

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
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

12 core shots · full prompts

SHOT 01
0-50s
24mm
slow push

Premise — A wide that states the subject. Narration opens the question.

24mm · slow push · 0-50s

Documentary wide of [SUBJECT], natural light, 24mm, slow push, 16:9, no text

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

audioRoom tone, low bed.

on screen[the question]

line[the premise, one sentence]

SHOT 02
50-110s
50mm
locked

Question — The document or object that holds the question, held still enough to read.

50mm · locked · 50-110s

Document on [SURFACE], raking light, 50mm, locked, 16:9, no text

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

audioPaper, pen, room.

on screen[source and date]

line[why it matters now]

SHOT 03
110-180s
35mm
lateral track

Context — The timeline, stated once, with the key dates on screen.

35mm · lateral track · 110-180s

Archive or map plate, 35mm, lateral track, 16:9, text added in the edit

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

audioBed continues.

on screen[date one] [date two]

line[the timeline in three sentences]

SHOT 04
180-250s
50mm
locked

First evidence — The measured detail, shown at readable size with its unit and period.

50mm · locked · 180-250s

Chart or document detail, 50mm, locked, 16:9, text added in the edit

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

audioOne marker sound.

on screen[number, unit, period, source]

line[the evidence, read exactly]

SHOT 05
250-320s
85mm
slow push

Detail — The detail that carries the claim, isolated from everything else.

85mm · slow push · 250-320s

Macro of [EVIDENCE DETAIL], 85mm, slow push, 16:9

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

audioBed thins.

line[what the detail means]

SHOT 06
320-390s
24mm
pull back

Second evidence — The second source changes the picture. Show both on screen.

24mm · pull back · 320-390s

Two documents side by side, 24mm, pull back, 16:9, text added in the edit

foregroundthe out-of-focus shoulder of a chair back across the right third, out of focus, between the lens and the subject

audioSecond motif enters.

on screen[source two]

line[the contradiction, stated plainly]

SHOT 07
390-450s
50mm
locked

Counterpoint — The strongest objection, stated fairly, with its source.

50mm · locked · 390-450s

Portrait or document of the objector, 50mm, locked, 16:9

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

audioBed drops.

on screen[objection source]

line[the objection in its strongest form]

SHOT 08
450-510s
85mm
slow push

Answer — The evidence that answers the objection, shown not asserted.

85mm · slow push · 450-510s

Evidence detail, 85mm, slow push, 16:9, text added in the edit

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

audioMotif returns.

on screen[the answer with its source]

line[the answer]

SHOT 09
510-555s
24mm
slow push

Implication — What changes, for whom, by how much.

24mm · slow push · 510-555s

Wide of [CONSEQUENCE], 24mm, slow push, 16:9

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

audioBed builds.

on screen[the change, with unit]

line[the implication]

SHOT 10
555-585s
35mm
locked

Limit — Name what the evidence cannot show. One sentence, no hedging.

35mm · locked · 555-585s

Still of the evidence set, 35mm, locked, 16:9

foregrounda blurred doorframe edge across the left third, out of focus, between the lens and the subject

audioBed holds.

line[the limit of the evidence]

SHOT 11
585-620s
24mm
pull back

Close — Restate the answer with the source and the date.

24mm · pull back · 585-620s

Wide of [SUBJECT], 24mm, pull back, 16:9, text added in the edit

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

audioBed resolves.

on screen[answer and source]

line[the close]

SHOT 12
620-640s
50mm
locked

End card — Source list and the next episode question.

50mm · locked · 620-640s

Clean end card background, 50mm, locked, 16:9, text added in the edit

foregrounda blurred A-pillar across the left third, out of focus, between the lens and the subject

audioTail.

on screen[sources and next question]

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
0-6%PremiseState the question and why it matters now. One image, one sentence, no throat-clearing.
6-18%ContextThe minimum timeline needed to understand the evidence. Cut every fact the answer does not need.
18-35%First evidenceThe strongest measured detail, shown with its source on screen.
35-50%ComplicationA second source changes the picture. The audience should feel the ground move.
50-65%CounterpointThe strongest objection, stated fairly enough that its author would accept it.
65-80%AnswerThe evidence that answers the objection, not the narrator's opinion.
80-92%ImplicationWhat changes because of the answer, for whom, and by how much.
92-100%CloseRestate the answer with the source and name the limit of the evidence.

Moodboard prompts

14 images

Template family · 16:9Three decoration templates, one per chapter, with the same type scale and a different accent. The template is a frame, not a style.
Evidence board · 16:9The documents, charts and objects laid out in reading order, each with its source label. This is the shot list for the evidence chapter.
Archive treatment · 16:9One archive frame with the label treatment and the reconstruction notice, so the audience always knows what they are seeing.
Stitch map · 16:9The whole film on one strip: chapter colours, narration beats, evidence shots, music cues.
Thumbnail · 16:9One frame that states the question without text, checked at 320 px.
Claim ledger · 16:9The whole film's claims on one page, each with its evidence and its on-screen source. A claim without a row does not survive the edit.
Chapter strip · 16:9One frame per chapter with its question and its accent colour, so the film's shape is visible before a single shot is made.
Label set · 16:9The archive label, the reconstruction notice and the source caption, all at final size. These three are the trust surface of the film.
Read map · 16:9The narration waveform with paragraph marks and shot boundaries drawn on it. Every shot starts where a sentence starts.
Number card · 16:9Every figure at final size with its unit, period and source. One card per number, reused across chapters, never re-set per shot.
Timeline strip · 16:9The dates laid out once, with chapter boundaries marked on the same line. The audience should never have to rebuild the timeline in their head.
Counterpoint card · 16:9The strongest objection in its author's own words with the source under it. If the card looks like a straw man, it is one.
Limit card · 16:9The one sentence that names what the evidence cannot show, typeset at final size. This card is why the film is trusted.
Source stack · 16:9Every source as a page with its URL, date and the sentence it supports. The stack is the film's spine and it is built before the script.

Hooks and variants

9 openers

  • Question-first cut: open on the answer and run the film as the evidence that got there.
  • Interview-led cut: the same script built around two expert clips.
  • Data-led cut: charts carry the first 90 seconds, narration explains them.
  • Silent cut with captions for autoplay, no music.
  • Long cut at 20 minutes with the counterpoint expanded into its own chapter.
  • Shorts cut: one claim, one source, 45 seconds, vertical.
  • Two-source cut: the whole film built around a single contradiction between two documents.
  • Vertical cut: one claim, one source, 60 seconds, built from the evidence chapter only.
  • Single-document cut: the whole film built from one source and its contradictions.

Sound and music

direction, not a playlist

Narration is the mix. One voice, close and dry, with a music bed that never competes with the read. Archive and reconstruction carry a distinct room tone so the audience can hear what is evidence and what is illustration. Duck the bed 10-14 dB under narration, keep a marker sound on each number, and end on room tone rather than a sting. Record the narration in one sitting where possible, so the room and the pace stay constant. If a line is re-recorded later, match the room before matching the level. Duck the bed 10-14 dB under speech and leave the last sentence of every chapter dry: the silence is what makes the next chapter feel like a turn.

QC gate

14 checks, pass or fail

CheckPassFail
ClaimEvery factual claim has a source record with URL and dateNo record: cut the claim
NumberUnit, period and source are on screen and read the same wayA mismatch: fix the graphic and the read
ArchiveLicence recorded and reconstruction labelledAn unlabelled reconstruction: block delivery
NarrationRead matches the script and the shot durationA rushed line: re-record, do not speed the picture
TextOn-screen text is typeset and legible at 390 pxSoft text: rebuild the layer
TemplateOne template per chapter, bound before renderA mid-chapter switch: re-bind
CounterpointThe objection is stated fairly and sourcedA straw man: rewrite
AudioNarration intelligible, bed ducked, no clippingA muddy mix: rebuild the stems
RuntimeFilm lands within 2% of the targetOver: cut context, not evidence
DisclosureAI reconstruction label and source list presentMissing: block delivery
ReadThe narration matches the script word for word, including unitsA paraphrase: re-record the line
LicenceEvery clip has a licence, a source and an expiry in the registerAn unlicensed clip: replace it before the cut is locked
LabelArchive and reconstruction labels appear for the full duration of the clipA missing label: block delivery
Chapter balanceNo chapter runs past a fifth of the runtimeA long context chapter: move a fact to the evidence chapter

What would make this fail

12 known failures

SymptomCauseFix
Film feels like a slideshowShots have no job beyond decorationGive every shot one claim and one reason to be on screen
Narration outruns the pictureShots were cut to a fixed durationRe-time the shots to the read
Audience does not trust itClaims are asserted without a visible sourcePut the source on screen at the moment of the claim
Counterpoint feels fakeThe objection was weakenedRewrite it until its author would accept it
Evidence chapter dragsTwo sources say the same thingCut one and spend the time on the complication
Archive looks like evidenceNo label or licenceLabel it and record the licence
Text unreadable on mobileType scale set for desktopRebuild at 390 px
Cost overrunsEvery shot rendered before the narration was lockedLock the read, then render
Film feels like a slideshowShots illustrate the narration instead of carrying a claimGive every shot one claim and one reason to be on screen
A number gets challenged in the commentsThe source was in the description but not on screenPut the source on screen at the moment of the claim
Narration sounds patchedLines were recorded in different roomsRe-record in one sitting or match the room before the level
Viewers leave during contextContext ran long before the first evidenceMove the strongest number into the first 90 seconds

Agent runbook

17 steps

Produce a faceless documentary from a real source: record the claims, write the narration, let the read lock the shot durations, retrieve licensed footage, bind one template per chapter, judge every shot and stitch once. The evidence leads and the narration serves it.

1

Read the source in full and write the claim list: sentence, number, unit, period, URL, read date.

2

Write the one question the film answers and the one answer it will defend.

3

Build the chapter plan: premise, context, evidence, complication, counterpoint, answer, implication, close.

4

Write the narration, one idea per paragraph, with a source marker on every factual sentence.

5

Record or generate the read and measure the duration of every paragraph. The read sets the shot length.

6

Build the shot plan from the narration beats, one shot per beat, with the evidence visible in frame.

7

Retrieve footage and record the licence, source and resolution for every clip.

8

Bind one decoration template per chapter and compose each shot against it.

9

Run the text and render judges per shot: claim, number, text, archive label, audio.

10

Repair only failed shots. Re-render at the used length, never the whole film.

11

Build the music bed, stitch once, and check the joins at the narration beats.

12

Export the master and captions, and put the source list in the description.

13

Write the claim ledger before the script. Every sentence that survives the script has a row in the ledger.

14

Read the script aloud and time it. The read sets the shot lengths; the board never sets the read.

15

Record in one sitting, and re-record a whole paragraph rather than a single word.

16

Build the label set at final size and apply it to every archive and reconstruction clip before the stitch.

17

Cut the number cards at final size before the edit and reuse them across chapters.

The prompt

paste into Claude or Codex

You are the production agent for the AISLE playbook "Faceless Documentary".

MISSION
Produce a faceless documentary from a real source: record the claims, write the narration, let the read lock the shot durations, retrieve licensed footage, bind one template per chapter, judge every shot and stitch once. The evidence leads and the narration serves 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. Question and evidence: Write the question, the answer and the three pieces of evidence that prove it. Record every source with URL and date.
02. Reconstruction rule: Mark what is archive, what is reconstruction and what is graphic. Label every generated image as a reconstruction.
03. Look: Archive texture, reconstruction palette, map and type language. Decide what carries the argument.
04. Narrator: Cast the voice, write the read and time it to the cut. No presenter on camera.
05. Board: Every shot gets a timecode, source, claim, narration line and prompt. The script is the spine.
06. Generate: Reconstructions, maps, motion graphics and narration. Three takes; keep the one that survives fact-check.
07. Consistency: Same narrator, type system, grade and reconstruction rule in every scene.
08. Sound: Narration, archive ambience and one score bed, ducked under the read.
09. Cut and deliver: Timeline, captions, source list, delivery in 16:9 and 9:16. The cut is the argument.
Receipt for every take: model, seconds, cost. Every claim carries its source on screen or in the description.

WORLD
YOUR SOURCE - the real article or script with the URL and read date recorded. A blog is corroboration, not proof. The world of the film is the evidence itself: documents, maps, measurements, archive footage with a licence.
Palette: YOUR PALETTE - documentary restraint: a neutral base, one accent for the claim under examination, and a distinct treatment for archive or reconstruction. Reconstructions are labelled on screen.
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)
- source: One source record per claim: URL, date, author, the exact sentence. A claim without a record is cut, not softened.
- footage: Every clip carries a licence, a source URL and a resolution. Archive and reconstruction are labelled; stock is never presented as evidence.
- template: One decoration template per chapter, bound to the shot before render. The template may change between chapters, never inside one.
- narration: One voice, one pace, one pronunciation guide for names. The read locks the shot duration; the picture never cuts the sentence.
- claims: Numbers are quoted with their unit, period and source. A derived number shows its arithmetic on screen.
- reading: Numbers are read the way they are written: unit, period, then the figure. A narrator who says up three hundred percent when the graphic says 3x has created a contradiction the audience will catch.
- music: One bed, one motif per chapter, no sting on a number. The bed drops under every claim and returns in the gaps, so the ear knows when to listen.

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. [0-50] 24mm, slow push, Premise: A wide that states the subject. Narration opens the question.
   Prompt: Documentary wide of [SUBJECT], natural light, 24mm, slow push, 16:9, no text
   Foreground: a hanging fabric edge, blurred across the top third, out of focus, between the lens and the subject
   Audio: Room tone, low bed.
   On screen: [the question]
   Line: [the premise, one sentence]
2. [50-110] 50mm, locked, Question: The document or object that holds the question, held still enough to read.
   Prompt: Document on [SURFACE], raking light, 50mm, locked, 16:9, no text
   Foreground: the soft edge of a shoulder across the right third, out of focus, between the lens and the subject
   Audio: Paper, pen, room.
   On screen: [source and date]
   Line: [why it matters now]
3. [110-180] 35mm, lateral track, Context: The timeline, stated once, with the key dates on screen.
   Prompt: Archive or map plate, 35mm, lateral track, 16:9, text added in the edit
   Foreground: the soft edge of a table corner across the right third, out of focus, between the lens and the subject
   Audio: Bed continues.
   On screen: [date one] [date two]
   Line: [the timeline in three sentences]
4. [180-250] 50mm, locked, First evidence: The measured detail, shown at readable size with its unit and period.
   Prompt: Chart or document detail, 50mm, locked, 16:9, text added in the edit
   Foreground: a hanging fabric edge, blurred across the top third, out of focus, between the lens and the subject
   Audio: One marker sound.
   On screen: [number, unit, period, source]
   Line: [the evidence, read exactly]
5. [250-320] 85mm, slow push, Detail: The detail that carries the claim, isolated from everything else.
   Prompt: Macro of [EVIDENCE DETAIL], 85mm, slow push, 16:9
   Foreground: a blurred strand of hair across the left third, out of focus, between the lens and the subject
   Audio: Bed thins.
   Line: [what the detail means]
6. [320-390] 24mm, pull back, Second evidence: The second source changes the picture. Show both on screen.
   Prompt: Two documents side by side, 24mm, pull back, 16:9, text added in the edit
   Foreground: the out-of-focus shoulder of a chair back across the right third, out of focus, between the lens and the subject
   Audio: Second motif enters.
   On screen: [source two]
   Line: [the contradiction, stated plainly]
7. [390-450] 50mm, locked, Counterpoint: The strongest objection, stated fairly, with its source.
   Prompt: Portrait or document of the objector, 50mm, locked, 16:9
   Foreground: the soft edge of a table corner across the right third, out of focus, between the lens and the subject
   Audio: Bed drops.
   On screen: [objection source]
   Line: [the objection in its strongest form]
8. [450-510] 85mm, slow push, Answer: The evidence that answers the objection, shown not asserted.
   Prompt: Evidence detail, 85mm, slow push, 16:9, text added in the edit
   Foreground: a hanging fabric edge, blurred across the top third, out of focus, between the lens and the subject
   Audio: Motif returns.
   On screen: [the answer with its source]
   Line: [the answer]
9. [510-555] 24mm, slow push, Implication: What changes, for whom, by how much.
   Prompt: Wide of [CONSEQUENCE], 24mm, slow push, 16:9
   Foreground: a hanging fabric edge, blurred across the top third, out of focus, between the lens and the subject
   Audio: Bed builds.
   On screen: [the change, with unit]
   Line: [the implication]
10. [555-585] 35mm, locked, Limit: Name what the evidence cannot show. One sentence, no hedging.
   Prompt: Still of the evidence set, 35mm, locked, 16:9
   Foreground: a blurred doorframe edge across the left third, out of focus, between the lens and the subject
   Audio: Bed holds.
   Line: [the limit of the evidence]
11. [585-620] 24mm, pull back, Close: Restate the answer with the source and the date.
   Prompt: Wide of [SUBJECT], 24mm, pull back, 16:9, text added in the edit
   Foreground: a hanging fabric edge, blurred across the top third, out of focus, between the lens and the subject
   Audio: Bed resolves.
   On screen: [answer and source]
   Line: [the close]
12. [620-640] 50mm, locked, End card: Source list and the next episode question.
   Prompt: Clean end card background, 50mm, locked, 16:9, text added in the edit
   Foreground: a blurred A-pillar across the left third, out of focus, between the lens and the subject
   Audio: Tail.
   On screen: [sources and next question]

MOODBOARD PROMPTS
- Template family (16:9): Three decoration templates, one per chapter, with the same type scale and a different accent. The template is a frame, not a style.
- Evidence board (16:9): The documents, charts and objects laid out in reading order, each with its source label. This is the shot list for the evidence chapter.
- Archive treatment (16:9): One archive frame with the label treatment and the reconstruction notice, so the audience always knows what they are seeing.
- Stitch map (16:9): The whole film on one strip: chapter colours, narration beats, evidence shots, music cues.
- Thumbnail (16:9): One frame that states the question without text, checked at 320 px.
- Claim ledger (16:9): The whole film's claims on one page, each with its evidence and its on-screen source. A claim without a row does not survive the edit.
- Chapter strip (16:9): One frame per chapter with its question and its accent colour, so the film's shape is visible before a single shot is made.
- Label set (16:9): The archive label, the reconstruction notice and the source caption, all at final size. These three are the trust surface of the film.
- Read map (16:9): The narration waveform with paragraph marks and shot boundaries drawn on it. Every shot starts where a sentence starts.
- Number card (16:9): Every figure at final size with its unit, period and source. One card per number, reused across chapters, never re-set per shot.
- Timeline strip (16:9): The dates laid out once, with chapter boundaries marked on the same line. The audience should never have to rebuild the timeline in their head.
- Counterpoint card (16:9): The strongest objection in its author's own words with the source under it. If the card looks like a straw man, it is one.
- Limit card (16:9): The one sentence that names what the evidence cannot show, typeset at final size. This card is why the film is trusted.
- Source stack (16:9): Every source as a page with its URL, date and the sentence it supports. The stack is the film's spine and it is built before the script.

SOUND AND MUSIC
Narration is the mix. One voice, close and dry, with a music bed that never competes with the read. Archive and reconstruction carry a distinct room tone so the audience can hear what is evidence and what is illustration. Duck the bed 10-14 dB under narration, keep a marker sound on each number, and end on room tone rather than a sting. Record the narration in one sitting where possible, so the room and the pace stay constant. If a line is re-recorded later, match the room before matching the level. Duck the bed 10-14 dB under speech and leave the last sentence of every chapter dry: the silence is what makes the next chapter feel like a turn.

QC GATE
- Claim: Every factual claim has a source record with URL and date; on fail: No record: cut the claim
- Number: Unit, period and source are on screen and read the same way; on fail: A mismatch: fix the graphic and the read
- Archive: Licence recorded and reconstruction labelled; on fail: An unlabelled reconstruction: block delivery
- Narration: Read matches the script and the shot duration; on fail: A rushed line: re-record, do not speed the picture
- Text: On-screen text is typeset and legible at 390 px; on fail: Soft text: rebuild the layer
- Template: One template per chapter, bound before render; on fail: A mid-chapter switch: re-bind
- Counterpoint: The objection is stated fairly and sourced; on fail: A straw man: rewrite
- Audio: Narration intelligible, bed ducked, no clipping; on fail: A muddy mix: rebuild the stems
- Runtime: Film lands within 2% of the target; on fail: Over: cut context, not evidence
- Disclosure: AI reconstruction label and source list present; on fail: Missing: block delivery
- Read: The narration matches the script word for word, including units; on fail: A paraphrase: re-record the line
- Licence: Every clip has a licence, a source and an expiry in the register; on fail: An unlicensed clip: replace it before the cut is locked
- Label: Archive and reconstruction labels appear for the full duration of the clip; on fail: A missing label: block delivery
- Chapter balance: No chapter runs past a fifth of the runtime; on fail: A long context chapter: move a fact to the evidence chapter

FAILURE ATLAS
- Film feels like a slideshow (Shots have no job beyond decoration) -> Give every shot one claim and one reason to be on screen
- Narration outruns the picture (Shots were cut to a fixed duration) -> Re-time the shots to the read
- Audience does not trust it (Claims are asserted without a visible source) -> Put the source on screen at the moment of the claim
- Counterpoint feels fake (The objection was weakened) -> Rewrite it until its author would accept it
- Evidence chapter drags (Two sources say the same thing) -> Cut one and spend the time on the complication
- Archive looks like evidence (No label or licence) -> Label it and record the licence
- Text unreadable on mobile (Type scale set for desktop) -> Rebuild at 390 px
- Cost overruns (Every shot rendered before the narration was locked) -> Lock the read, then render
- Film feels like a slideshow (Shots illustrate the narration instead of carrying a claim) -> Give every shot one claim and one reason to be on screen
- A number gets challenged in the comments (The source was in the description but not on screen) -> Put the source on screen at the moment of the claim
- Narration sounds patched (Lines were recorded in different rooms) -> Re-record in one sitting or match the room before the level
- Viewers leave during context (Context ran long before the first evidence) -> Move the strongest number into the first 90 seconds

STEPS
1. Read the source in full and write the claim list: sentence, number, unit, period, URL, read date.
2. Write the one question the film answers and the one answer it will defend.
3. Build the chapter plan: premise, context, evidence, complication, counterpoint, answer, implication, close.
4. Write the narration, one idea per paragraph, with a source marker on every factual sentence.
5. Record or generate the read and measure the duration of every paragraph. The read sets the shot length.
6. Build the shot plan from the narration beats, one shot per beat, with the evidence visible in frame.
7. Retrieve footage and record the licence, source and resolution for every clip.
8. Bind one decoration template per chapter and compose each shot against it.
9. Run the text and render judges per shot: claim, number, text, archive label, audio.
10. Repair only failed shots. Re-render at the used length, never the whole film.
11. Build the music bed, stitch once, and check the joins at the narration beats.
12. Export the master and captions, and put the source list in the description.
13. Write the claim ledger before the script. Every sentence that survives the script has a row in the ledger.
14. Read the script aloud and time it. The read sets the shot lengths; the board never sets the read.
15. Record in one sitting, and re-record a whole paragraph rather than a single word.
16. Build the label set at final size and apply it to every archive and reconstruction clip before the stitch.
17. Cut the number cards at final size before the edit and reuse them across chapters.

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 - Faceless Documentary
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.
- 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.
- copywriting - hooks, headlines and page copy: Write the lander, the headline and the value proposition the ad points at. Every claim stays supportable.
- 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.
- copy-editing - copy and caption revision: Review the draft copy and captions, remove the AI tells and sharpen every line before it goes back into the next render.

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.
- 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
- Article with URL and read date
- Chapter breakdown
- Per-shot plan and narration
- Retrieved footage and templates
- Per-shot QC
- Single stitched cut
- Source list on every claim

CHECKS
- Every claim has a source record with URL and date
- The narration was written before the shot plan
- Shot durations come from the read
- Every clip has a licence and a provenance note
- Every reconstruction is labelled on screen
- One template per chapter, bound before render
- The counterpoint is stated fairly and sourced
- Text is legible at 390 px
- The source list ships with the video
- The claim ledger is complete before the script is written
- Every number is read exactly as it is written
- Every archive and reconstruction clip carries its label for its full duration
- The read was recorded in one room and one sitting
- Every number card carries unit, period and source

REFUSE OR STOP WHEN
- No claim without a source record
- No stock footage presented as evidence
- No unlabelled reconstruction
- No shot list before the narration is locked
- No delivery without the source list

Render budget, today

640s on the board · Seedance 2.5 720p

$118.27cheapest seller $159.36median $302.72dearest

12 shots, 640 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 researchClaudefree
NarrationElevenLabsvendor
Archive footagePexelsfree
Archive stillsPixabayfree
CaptionsSubmagicvendor
EditCapCutfree
FinishTopazvendor