Studio · Cartoon · 9:16 · 3-7s fragments
AI Cartoon Short
A repeatable cartoon pipeline: locked model sheets, a fragmented timeline of 3-7 second shots, a four-layer sound plan and cross-episode memory. The medium is locked once so a flashback cannot contaminate the look, and every short ends on a hook that the next short answers.
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
10 deliverables
- Series bible: medium, palette, cast and the rule of the world
- Model sheet per character, front and three-quarter, with a paste-ready lock
- Fragment timeline: 3-7 second shots with a job each
- Four-layer sound plan: voice, effects, music, room
- Shot board with action, on-screen text and line
- Cross-episode memory file: what changed, what must not change
- QC report on identity, medium, timing and the end hook
- Prop sheet: every recurring prop drawn once with its fixed hand, size and wear
- Caption card: one font, one weight, one position, checked at 390 px
- Stem map: voice, effects, music and room delivered as four separate files
World and look
paste before anything
| World | MULTIPLE ANIMATION - one medium locked. Keep the style anchor on the medium only so a flashback cannot contaminate the look. State the medium in one sentence and paste it into every prompt. |
| Palette | YOUR PALETTE - fixed base, subject and accent layers, identical wording across episodes. A new episode may add one prop colour, never a new palette. |
| 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
| location | One card per set: layout, door, window, the one prop the camera returns to. A flashback uses the same set with one changed detail, not a new set. |
| props | Every recurring prop gets a card and a fixed hand. If a prop changes size between shots, the short loses its world. |
| wardrobe | Cartoon wardrobe is part of the design, not a costume change. If a character changes clothes, the model sheet changes with it. |
| medium | One medium sentence, pasted verbatim: line weight, shading model, background treatment. Never mix a 3D look into a 2D short. |
| voice | One voice per character, fixed pace and one verbal habit. The same voice library carries across episodes. |
| colour | Three named swatches per episode: base, subject, accent. The accent belongs to one character only, so a silhouette reads as that character before the face does. |
| hands | Every character has a fixed hand shape and a fixed finger count for the medium. Hands are where cartoon identity fails first, so they are drawn on the model sheet and referenced, never described. |
Every lock is pasted into the prompt for every shot that touches it. A lock that is paraphrased is not a lock.
Cast locks
2 identities
CAST A
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.
CAST B
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
4 skills, in order
| Skill | Job in this run |
|---|---|
copywriting | hooks, headlines and page copy |
image | keyframes, model sheets and moodboards |
video | shots, motion, editing and the cut |
social | organic and platform format |
The SKILL.md installs these in order. Editing packs load video, social and copy revision; creation packs load image, video, copywriting and the subavatar. A pack never loads a skill it does not use.
Ad sources this pack reads
plug in your library
| Source | What it gives |
|---|---|
| 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
0.0-4.0s
24mm
slow push
Establish — [CAST A] in the locked world, medium fixed, no dialogue.
[CAST A LOCK] in [SET CARD], [MEDIUM SENTENCE], 24mm, slow push, vertical 9:16, no text
foregroundthe out-of-focus dash edge across the bottom third, out of focus, between the lens and the subject
audioRoom tone and one signature sound.
4.0-9.0s
35mm
locked
Incident — An event breaks the normal; [CAST B] reacts first.
[CAST B LOCK] reacting in [SET CARD], event in foreground, [MEDIUM SENTENCE], 35mm, locked, vertical 9:16
foregroundthe out-of-focus dash edge across the bottom third, out of focus, between the lens and the subject
audioImpact, then a comic beat.
line[one line from CAST B]
9.0-14.0s
50mm
slow push
Escalate — [CAST A] tries the obvious fix and makes it worse.
[CAST A LOCK] attempting [PROP CARD], consequence visible, [MEDIUM SENTENCE], 50mm, slow push, vertical 9:16
foregrounda passing guard rail, smeared by motion across the right third, out of focus, between the lens and the subject
audioRising effect, one wrong note.
on screen[the obvious fix, named in three words]
14.0-19.0s
85mm
locked
Reaction — The consequence lands on a face. No cut away.
[CAST A LOCK] close-up, held reaction, [MEDIUM SENTENCE], 85mm, locked, vertical 9:16
foregrounda blurred strand of hair across the left third, out of focus, between the lens and the subject
audioSilence under the face.
line[one short line]
19.0-24.0s
35mm
whip pan
Turn — [CAST A] chooses the unexpected move; the camera follows late.
[CAST A LOCK] moving fast through [SET CARD], [MEDIUM SENTENCE], 35mm, whip pan, motion blur, vertical 9:16
foregrounda blurred A-pillar across the left third, out of focus, between the lens and the subject
audioWhoosh, music enters.
on screen[the choice, named in four words]
24.0-30.0s
50mm
slow push
Payoff — The choice works, but costs something visible: a crack, a missing piece, a stain.
[CAST A LOCK] and [PROP CARD] after the choice, damage visible, [MEDIUM SENTENCE], 50mm, slow push, vertical 9:16
foregrounda blurred monitor edge across the left third, out of focus, between the lens and the subject
audioMusic resolves, one small crack.
on screen[the cost, named in three words]
line[the cost named in one line]
30.0-35.0s
24mm
pull back
Button — The world returns, changed by one detail the audience can spot.
Wide of [SET CARD], one changed detail, [CAST A LOCK] small in frame, [MEDIUM SENTENCE], 24mm, pull back, vertical 9:16
foregroundthe out-of-focus dash edge across the bottom third, out of focus, between the lens and the subject
audioRoom tone returns.
35.0-40.0s
85mm
locked
Next-hook — A single frame teases the next short: a new object, a shadow, a door opening.
Single frame, [NEXT OBJECT] in [SET CARD], [MEDIUM SENTENCE], 85mm, locked, vertical 9:16, no text
foregroundthe near edge of a wall corner across the bottom third, out of focus, between the lens and the subject
audioOne sound from the next short.
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 |
|---|---|---|
| 0-7% | Establish | World and cast in one image, medium fixed, no dialogue yet. |
| 7-30% | Incident | An event breaks the normal. The audience sees it before the cast does. |
| 30-45% | Escalate | The obvious fix makes it worse in a visual way. |
| 45-60% | Reaction | The consequence lands on a face. Hold the frame. |
| 60-75% | Turn | A choice that the audience did not expect but the character would make. |
| 75-88% | Payoff | The choice works and costs something visible. |
| 88-96% | Button | The world returns, changed by one detail. |
| 96-100% | Next-hook | One frame teases the next short without explaining it. |
Moodboard prompts
14 images
Hooks and variants
10 openers
- Silent episode with captions only, built for autoplay without sound.
- Swap the cast roles: the reactor becomes the cause.
- Tell the same 40 seconds from the prop's point of view.
- Open on the next-hook frame and run the episode as the explanation.
- Two-episode arc cut as one 80-second short for YouTube.
- Black-and-white flashback using the same set and one changed detail.
- One-character episode: the whole short carried by a single performance and a prop.
- No-dialogue episode: expressions carry the story and captions carry the joke.
- Prop-led episode: the character never speaks and the prop carries every beat.
- Two-short arc: the same forty seconds split across two uploads, each ending on a different hook.
Sound and music
direction, not a playlist
Four layers, always: voice dry and close, effects exaggerated for the medium, music one loop that returns every episode, room tone under everything. Cartoon sound is allowed to be bigger than the picture. Keep the voice and music on separate stems so a platform cut can duck the loop without losing the line. Deliver four stems, not one mix: voice, effects, music, room. Name them scene-number-role so a re-cut can drop the music and keep the line. Effects are drawn rather than sampled: a boing, a pop, a slide whistle, each owned by one character and reused as their signature.
QC gate
14 checks, pass or fail
| Check | Pass | Fail |
|---|---|---|
| Identity | Model sheet matches in every shot | Drift: regenerate from the sheet |
| Medium | The style sentence is identical in every prompt | A mixed medium: discard the take |
| Fragment length | Every shot lands between 3 and 7 seconds | Outside the band: re-cut or re-board |
| Prop continuity | Recurring props keep size, colour and hand | A changed prop: regenerate |
| Set continuity | The set keeps its layout and one light source | A new wall: restore the plate |
| Voice | One voice per character, same pace | A swapped voice: re-voice |
| Hook | The last frame teases the next short without explaining it | A flat ending: rebuild the button |
| Sound | Four layers present and separable | A missing layer: rebuild the stem |
| Disclosure | AI label on the master | Missing: block delivery |
| Hands | Hand shape and finger count match the model sheet in every shot | A fused or extra finger: regenerate the shot |
| Caption | Caption style, position and size match the card and clear the platform UI | Overlap or drift: rebuild the caption layer |
| Runtime | The short lands between 35 and 45 seconds | Over: cut a fragment, never the reaction |
| Prop scale | The prop matches its sheet against the character's hand | A resized prop: regenerate from the sheet |
| Accent | The accent colour appears only on the character who owns it | A second character in the accent: re-grade |
What would make this fail
12 known failures
| Symptom | Cause | Fix |
|---|---|---|
| The look changes between episodes | The medium sentence was paraphrased | Paste the sentence verbatim and re-render the opener |
| The short feels slow | Fragments are longer than 7 seconds | Split the fragment and cut on the action |
| The cast reads as different characters | No model sheet, or a sheet per shot | One sheet per character, reused |
| The joke does not land | The reaction shot is missing or too short | Hold the face one beat longer |
| The next hook confuses | It explains instead of teasing | Cut to the object, not the answer |
| Sound feels thin | Only music and voice were used | Add effects and room tone |
| Cost climbs across episodes | Every episode re-generates the cast | Reuse the sheet and the accepted seeds |
| Hands melt in motion | The hand was described in words instead of referenced from the sheet | Attach the hand reference and regenerate only that shot |
| Episode four looks like a different show | A new palette or line weight entered without a bible change | Restore the bible sentence and re-render the opener |
| Captions fight the picture | Caption style was improvised per episode | Apply the caption card and re-export |
| Every episode reads as a pilot | No running memory file was written | Write the memory file and read it before boarding |
| A fragment has no job | It was added for rhythm rather than story | Delete it or give it a job |
Agent runbook
18 steps
Run a cartoon short as a repeatable line: one medium, one sheet per character, 3-7 second fragments, four sound layers and a hook that carries to the next episode. Nothing about the look may change between episodes without a written reason.
Write the series bible in one page: medium sentence, palette, cast, the rule of the world and the running hook.
Draw one model sheet per character and one set plate per location. Approve them before any shot is generated.
Break the story into 3-7 second fragments with one job each. Delete any fragment that repeats another.
Write the action, on-screen text and line for every fragment.
Generate the first fragment and check identity, medium and prop before making the rest.
Generate the remaining fragments in small batches, keeping the sheet and the seed for each character.
Assemble in order and watch once without stopping. Note the fragments that fail, not the taste you dislike.
Re-generate only failed fragments. Never re-render a passing fragment to fix a neighbour.
Build the four sound layers, then mix with voice on top and music ducked under it.
Add the end hook and check that it teases without explaining.
Write the cross-episode memory: what changed, what must not change, which seeds were accepted.
Export the master, captions and AI label, and file the memory with the project.
Write the prop sheet and the expression sheet before boarding. A reaction that is not on the sheet does not exist.
Cut a silent pass first: picture and captions only, no music. If the story does not read silent, sound will not save it.
Score the episode after the silent pass locks: one loop, one signature sound per character, room tone under everything.
File the accepted seeds, the prop sheet and the caption card with the episode so the next short starts from the same assets.
Cut the fragment board before generating: first, middle and last frame for every fragment.
Check the loop frame against the next short's opener before the episode is exported.
The prompt
paste into Claude or Codex
You are the production agent for the AISLE playbook "AI Cartoon Short".
MISSION
Run a cartoon short as a repeatable line: one medium, one sheet per character, 3-7 second fragments, four sound layers and a hook that carries to the next episode. Nothing about the look may change between episodes without a written reason.
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. Script and gag: Write the setup, the turn and the button in one sentence each. If the button does not land, no prompt saves it.
02. Model sheets: Three original characters: turnarounds, expressions, wardrobe, line weight. No stock face, no borrowed likeness.
03. Look: Palette, line weight and background language. Lock the style before the first frame.
04. Board: Every shot gets a timecode, camera, action, expression and prompt. Direct the gag, not the vibe.
05. Generate: Keyframes, image-to-video, voices and SFX. Three takes; keep the one that reads at thumbnail size.
06. Consistency: Same model sheet, line weight and palette in every cut. Paste the lock and reuse the seed.
07. Sound: Voices, foley and one music bed, ducked under dialogue.
08. Cut and deliver: Timeline, captions, delivery in 9:16, 16:9 and 1:1. The cut is the product.
Receipt for every take: model, seconds, cost. The model sheet is the contract; a drifted line means a reshoot.
WORLD
MULTIPLE ANIMATION - one medium locked. Keep the style anchor on the medium only so a flashback cannot contaminate the look. State the medium in one sentence and paste it into every prompt.
Palette: YOUR PALETTE - fixed base, subject and accent layers, identical wording across episodes. A new episode may add one prop colour, never a new palette.
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)
- CAST A (Identity slot): CAST A: [fixed design], [fixed palette], [fixed proportions] - paste verbatim in every shot.
- CAST B (Identity slot): CAST B: [fixed design], [fixed palette], [fixed proportions] - paste verbatim in every shot.
LOCKS (location, props, wardrobe, axis)
- location: One card per set: layout, door, window, the one prop the camera returns to. A flashback uses the same set with one changed detail, not a new set.
- props: Every recurring prop gets a card and a fixed hand. If a prop changes size between shots, the short loses its world.
- wardrobe: Cartoon wardrobe is part of the design, not a costume change. If a character changes clothes, the model sheet changes with it.
- medium: One medium sentence, pasted verbatim: line weight, shading model, background treatment. Never mix a 3D look into a 2D short.
- voice: One voice per character, fixed pace and one verbal habit. The same voice library carries across episodes.
- colour: Three named swatches per episode: base, subject, accent. The accent belongs to one character only, so a silhouette reads as that character before the face does.
- hands: Every character has a fixed hand shape and a fixed finger count for the medium. Hands are where cartoon identity fails first, so they are drawn on the model sheet and referenced, never described.
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. [0.0-4.0] 24mm, slow push, Establish: [CAST A] in the locked world, medium fixed, no dialogue.
Prompt: [CAST A LOCK] in [SET CARD], [MEDIUM SENTENCE], 24mm, slow push, vertical 9:16, no text
Foreground: the out-of-focus dash edge across the bottom third, out of focus, between the lens and the subject
Audio: Room tone and one signature sound.
2. [4.0-9.0] 35mm, locked, Incident: An event breaks the normal; [CAST B] reacts first.
Prompt: [CAST B LOCK] reacting in [SET CARD], event in foreground, [MEDIUM SENTENCE], 35mm, locked, vertical 9:16
Foreground: the out-of-focus dash edge across the bottom third, out of focus, between the lens and the subject
Audio: Impact, then a comic beat.
Line: [one line from CAST B]
3. [9.0-14.0] 50mm, slow push, Escalate: [CAST A] tries the obvious fix and makes it worse.
Prompt: [CAST A LOCK] attempting [PROP CARD], consequence visible, [MEDIUM SENTENCE], 50mm, slow push, vertical 9:16
Foreground: a passing guard rail, smeared by motion across the right third, out of focus, between the lens and the subject
Audio: Rising effect, one wrong note.
On screen: [the obvious fix, named in three words]
4. [14.0-19.0] 85mm, locked, Reaction: The consequence lands on a face. No cut away.
Prompt: [CAST A LOCK] close-up, held reaction, [MEDIUM SENTENCE], 85mm, locked, vertical 9:16
Foreground: a blurred strand of hair across the left third, out of focus, between the lens and the subject
Audio: Silence under the face.
Line: [one short line]
5. [19.0-24.0] 35mm, whip pan, Turn: [CAST A] chooses the unexpected move; the camera follows late.
Prompt: [CAST A LOCK] moving fast through [SET CARD], [MEDIUM SENTENCE], 35mm, whip pan, motion blur, vertical 9:16
Foreground: a blurred A-pillar across the left third, out of focus, between the lens and the subject
Audio: Whoosh, music enters.
On screen: [the choice, named in four words]
6. [24.0-30.0] 50mm, slow push, Payoff: The choice works, but costs something visible: a crack, a missing piece, a stain.
Prompt: [CAST A LOCK] and [PROP CARD] after the choice, damage visible, [MEDIUM SENTENCE], 50mm, slow push, vertical 9:16
Foreground: a blurred monitor edge across the left third, out of focus, between the lens and the subject
Audio: Music resolves, one small crack.
On screen: [the cost, named in three words]
Line: [the cost named in one line]
7. [30.0-35.0] 24mm, pull back, Button: The world returns, changed by one detail the audience can spot.
Prompt: Wide of [SET CARD], one changed detail, [CAST A LOCK] small in frame, [MEDIUM SENTENCE], 24mm, pull back, vertical 9:16
Foreground: the out-of-focus dash edge across the bottom third, out of focus, between the lens and the subject
Audio: Room tone returns.
8. [35.0-40.0] 85mm, locked, Next-hook: A single frame teases the next short: a new object, a shadow, a door opening.
Prompt: Single frame, [NEXT OBJECT] in [SET CARD], [MEDIUM SENTENCE], 85mm, locked, vertical 9:16, no text
Foreground: the near edge of a wall corner across the bottom third, out of focus, between the lens and the subject
Audio: One sound from the next short.
MOODBOARD PROMPTS
- Model sheet (3:4): Front and three-quarter view, neutral pose, fixed line weight, fixed palette. The sheet is the contract, not a concept.
- Style anchor (16:9): One frame that fixes the medium: line, shading, background treatment. Paste the same sentence into every prompt.
- Fragment strip (9:16): Four frames from one 3-7 second fragment, so the action can be read before it is generated.
- Set plate (16:9): The set with no cast, four walls and the one prop, used as the plate behind every angle.
- Next-hook frame (9:16): The final teaser frame: one new object, same world, no text.
- Prop sheet (3:4): Every recurring prop on one sheet: front, side, in hand, at the size it appears. Same line weight as the cast. A prop redrawn per shot is a continuity bug waiting to happen.
- Expression sheet (3:4): Six expressions per character: neutral, surprise, delight, dread, resolve, collapse. Same head shape and eye spacing. The reaction shot is chosen from this sheet, never improvised.
- Colour script (16:9): One strip per beat, each with its dominant colour named. The accent appears only where the story turns. This is the check that stops a new palette arriving in episode four.
- Caption card (9:16): The caption style applied over the busiest background in the episode, so legibility is judged against the worst case rather than a clean plate.
- Fragment board (9:16): One fragment broken into its first, middle and last frame, so the action is readable before a single second is generated. If three frames do not tell the beat, the fragment is wrong.
- Loop check (9:16): The last frame beside the first frame of the next short. The hook reads as a continuation of the same show, not the start of a new one.
- Prop in hand (3:4): The recurring prop held by the character who owns it, at the size it appears on screen. Hand and prop are checked together because that is where both drift.
- Set at night (16:9): The set under its second light state, same walls and same door. A night episode is a re-light of the same set, never a new build.
- Sound sheet (16:9): The four stems drawn as layers with the signature sound marked on the character it belongs to. Sound is boarded with the picture, not added after it.
SOUND AND MUSIC
Four layers, always: voice dry and close, effects exaggerated for the medium, music one loop that returns every episode, room tone under everything. Cartoon sound is allowed to be bigger than the picture. Keep the voice and music on separate stems so a platform cut can duck the loop without losing the line. Deliver four stems, not one mix: voice, effects, music, room. Name them scene-number-role so a re-cut can drop the music and keep the line. Effects are drawn rather than sampled: a boing, a pop, a slide whistle, each owned by one character and reused as their signature.
QC GATE
- Identity: Model sheet matches in every shot; on fail: Drift: regenerate from the sheet
- Medium: The style sentence is identical in every prompt; on fail: A mixed medium: discard the take
- Fragment length: Every shot lands between 3 and 7 seconds; on fail: Outside the band: re-cut or re-board
- Prop continuity: Recurring props keep size, colour and hand; on fail: A changed prop: regenerate
- Set continuity: The set keeps its layout and one light source; on fail: A new wall: restore the plate
- Voice: One voice per character, same pace; on fail: A swapped voice: re-voice
- Hook: The last frame teases the next short without explaining it; on fail: A flat ending: rebuild the button
- Sound: Four layers present and separable; on fail: A missing layer: rebuild the stem
- Disclosure: AI label on the master; on fail: Missing: block delivery
- Hands: Hand shape and finger count match the model sheet in every shot; on fail: A fused or extra finger: regenerate the shot
- Caption: Caption style, position and size match the card and clear the platform UI; on fail: Overlap or drift: rebuild the caption layer
- Runtime: The short lands between 35 and 45 seconds; on fail: Over: cut a fragment, never the reaction
- Prop scale: The prop matches its sheet against the character's hand; on fail: A resized prop: regenerate from the sheet
- Accent: The accent colour appears only on the character who owns it; on fail: A second character in the accent: re-grade
FAILURE ATLAS
- The look changes between episodes (The medium sentence was paraphrased) -> Paste the sentence verbatim and re-render the opener
- The short feels slow (Fragments are longer than 7 seconds) -> Split the fragment and cut on the action
- The cast reads as different characters (No model sheet, or a sheet per shot) -> One sheet per character, reused
- The joke does not land (The reaction shot is missing or too short) -> Hold the face one beat longer
- The next hook confuses (It explains instead of teasing) -> Cut to the object, not the answer
- Sound feels thin (Only music and voice were used) -> Add effects and room tone
- Cost climbs across episodes (Every episode re-generates the cast) -> Reuse the sheet and the accepted seeds
- Hands melt in motion (The hand was described in words instead of referenced from the sheet) -> Attach the hand reference and regenerate only that shot
- Episode four looks like a different show (A new palette or line weight entered without a bible change) -> Restore the bible sentence and re-render the opener
- Captions fight the picture (Caption style was improvised per episode) -> Apply the caption card and re-export
- Every episode reads as a pilot (No running memory file was written) -> Write the memory file and read it before boarding
- A fragment has no job (It was added for rhythm rather than story) -> Delete it or give it a job
STEPS
1. Write the series bible in one page: medium sentence, palette, cast, the rule of the world and the running hook.
2. Draw one model sheet per character and one set plate per location. Approve them before any shot is generated.
3. Break the story into 3-7 second fragments with one job each. Delete any fragment that repeats another.
4. Write the action, on-screen text and line for every fragment.
5. Generate the first fragment and check identity, medium and prop before making the rest.
6. Generate the remaining fragments in small batches, keeping the sheet and the seed for each character.
7. Assemble in order and watch once without stopping. Note the fragments that fail, not the taste you dislike.
8. Re-generate only failed fragments. Never re-render a passing fragment to fix a neighbour.
9. Build the four sound layers, then mix with voice on top and music ducked under it.
10. Add the end hook and check that it teases without explaining.
11. Write the cross-episode memory: what changed, what must not change, which seeds were accepted.
12. Export the master, captions and AI label, and file the memory with the project.
13. Write the prop sheet and the expression sheet before boarding. A reaction that is not on the sheet does not exist.
14. Cut a silent pass first: picture and captions only, no music. If the story does not read silent, sound will not save it.
15. Score the episode after the silent pass locks: one loop, one signature sound per character, room tone under everything.
16. File the accepted seeds, the prop sheet and the caption card with the episode so the next short starts from the same assets.
17. Cut the fragment board before generating: first, middle and last frame for every fragment.
18. Check the loop frame against the next short's opener before the episode is exported.
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 - AI Cartoon Short
These are the installed skills this run loads. Load them in this order and do not add a skill that is not on the list.
- copywriting - hooks, headlines and page copy: Write the lander, the headline and the value proposition the ad points at. Every claim stays supportable.
- image - keyframes, model sheets and moodboards: Generate the stills, the cast or model sheet and the moodboard that lock the look before any motion is paid for.
- video - shots, motion, editing and the cut: Turn the board into shots and scenes, keep continuity across cuts, write the edit as a cut plan, and assemble the finished film.
- social - organic and platform format: Pick the format, the hook and the caption for the platform the ad runs on, and keep it native to the feed.
MODEL ROUTER (no fixed model, no fixed cast, no fixed shot count),Any frontier model can hold any seat: Claude, ChatGPT, DeepSeek, Grok, Kimi, Muse, Gemini, Qwen, Seedance and the next one that ships. Use the model the buyer already pays for. The board, style.md, cast locks and receipts are portable between them.,The run is not a fixed number of shots or characters. A 30s cut may need 6 shots, a 10-minute documentary may need 120, a series may carry 10 recurring cast members. The core board is a starting point, not a cap. Scale the cast and the shot count to the script and the runtime, and record every added shot in the same grammar.,On a rate limit, outage or spend cap: save the run state (board, style.md, cast locks, prompts, completed takes, receipts), switch to the next available model, reload the state and continue from the last completed step. Never restart the run and never re-charge a completed take.,Language models own structure, copy, research and QC. Video models own motion and frames. Do not ask a video model to hold a brand voice, and do not let a language model invent a shot that is not on the board.
RUN ORDER
- INGEST: pull the ad from the source the buyer names. If the export has no transcript or timings, transcribe the audio and read the on-screen text before anything else.
- DECODE: write style.md. Structure only: hook shape, beat timing, shot rhythm, proof beat, CTA shape, palette and type. State what is copied and what is rebuilt.
- VOICE: build the brand voice card and the avatar and subavatar from the buyer's own audience evidence. Never reuse the source brand's voice.
- PLAN: write the board, the cut plan and the variant matrix. Price every take before it is generated.
- GENERATE: keyframes first, gate the first frame, then motion. Keep the cast, product and style locked between renders.
- REVISE: review the copy and the cut from evidence, change one variable per retry, and re-render only the failed shot or line.
- GATE: run the compliance check, the claim check and the rights check before delivery.
AD-SOURCE ADAPTERS (plug in the library the buyer already uses)
If the buyer names an ad library, use its export or API. If not, ask for one export, URL, screenshot, transcript or CSV. Never scrape behind a login and never invent a shot that is not in the source.
- 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
- Model sheets with fixed design and proportions
- Fragmented timeline sized in 3-7s shots
- Four-layer sound plan
- Cross-episode memory file
- Consistency check before the next short
CHECKS
- One medium sentence, pasted verbatim
- One model sheet per character, approved before generation
- Every fragment between 3 and 7 seconds
- Four sound layers present and separable
- The hook teases the next short
- The cross-episode memory is written
- The AI label ships with the master
- Every recurring prop has a sheet and keeps its size and hand
- The accent colour appears only at the turn
- The silent pass reads before any sound is added
- Four stems are delivered, named by scene and role
- The fragment board reads before any fragment is generated
- The last frame hands off to the next short
REFUSE OR STOP WHEN
- No shot before the model sheet is approved
- No mixed medium inside one series
- No fragment over 7 seconds
- No new palette per episode
- No delivery without the memory file
Render budget, today
40s on the board · Seedance 2.5 720p
8 shots, 40 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 direction | Claude |
| Model sheets | GPT Image |
| Animation | Runway |
| Voices | ElevenLabs |
| Music | Suno |
| Edit and captions | CapCut |
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 scriptUGC · 9:16 · 100 ads a dayUGC Ad Factory
One shoot day, a hundred ads, each written for a real buyer.
6 shots / 30s ad · 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