Workflow Bench
Library- Recipes
- 11
- Takes
- 0
- Running
- 0
- Street interview · ads flow
- Not run1Script
- Not run2Start frame
- Not run3Clip plan
- Not run4Video prompts
- Not run5Video clips · Seedance 2
- Not run6Check
- Not run7Final video
Ads dashboard, liveNot ready for social - Animated ad
- 1The briefBy hand
Formula: villain and hero Animation look: 3D Pixar-style, soft night light, warm and funny The product: Zebra CBD sleep gummies The person: a woman in her late 30s who cannot sleep The problem: she lies awake at 3am, mind racing, and drags through the next day The cause, if the formula has a villain: a racing mind and stress that keep her body switched on at night Failed fixes: melatonin (groggy next morning), counting sheep and sleep apps (the mind still races) How the product works: calms the switched-on body so sleep comes without the morning fog Proof and offer: (blank) Narrator: an outside voice, warm British woman, 40s, a little wry Length and shape: about 45 seconds, 9:16 End card line: Sleep, sorted.
- 2StoryboardClaude Opus
You are the director and storyboard artist of a short ANIMATED ad. Lay it out from the brief and the brand pack, following the formula the brief names. The formulas (pull every beat of the one named, in order; merge two beats into one scene when the length is short): 1 VILLAIN AND HERO: the relatable problem as the hook; reveal the villain (the cause as a monster); show how bad it is; every usual fix fails (each failed fix can be its own small character); the hero mechanism; the product revealed as the hero; the person transformed; proof; call to action. 2 VILLAIN TELLS THE STORY: open inside the villain's world; the villain brags, comfortable, and the viewer realises they are the house; how it survives in you; its one weakness named before the product appears; the product arrives, seen as panic from the villain's side; the villains dissolve and the person wakes up; offer. The villain narrates. 3 TALKING OBJECT: an object speaks in first person and must BE the pain. Either the thing the buyer already uses confesses its flaw then hands over to the product, or the product introduces itself (what it is, what it is NOT, its mechanism against the rival, proof, "pick me up"), or the body part with the problem narrates on the viewer's side, then turns worried when the ingredients arrive. Emotion first, mechanism second, product third. 4 PIXAR STORY (thin, one source): a character in a situation the viewer knows; a clear want; an obstacle; the product as the turn; a better ending. Faces react, nobody explains. One exaggerated trait. 5 STICK FIGURES (thin, one ad): a curiosity story that is not about the product, then the turn to the product, narrated, comic panels. 6 INSIDE THE BODY TIMELINE: a question about what happens inside the body; the mechanism shown inside the body; a timeline (Day 1, 3, 7, 14, 30) switching between the science and a human moment; climbing from biology to symptom to daily life to who the person becomes; guarantee, then the offer. Rules for every formula: - The villain is visually distinctive and slightly funny. Its face: smug while it wins, WORRIED or panicking the moment the hero arrives, then beaten. Never a villain that stays smug to the end. - The person who suffers is AWAKE and upset in the problem scenes (eyes open, frowning, rubbing their face), never peacefully asleep. - The problem and the failed fixes come first. The product arrives at the turn, never in the opening scene (the one exception: the product itself as the talking object). - Name the failed fixes and why they fail. What feeds the villain is what the product removes. - Characters DO things every few seconds (shove, block, vacuum, dissolve, run). Nobody stands and talks. Every scene is one clear moment readable with the sound off. - The places are fixed: the same room stays the same room in every scene set there (same furniture, light, colours). Give each place a precise look. Use as few places as the story allows. - Only facts in the brief and the brand pack. No health cures promised. - The narrator speaks over the scenes; characters do not speak on screen. Narration at most 2.5 words a second of the scene. - Scenes are 4, 6, 8 or 10 seconds, 3 to 13 scenes, at most 120 seconds. At most 6 characters (the person, the villains, the failed fixes, the hero) and 4 places. The product counts as the hero only if it is drawn as a character; otherwise it is the real product, shown with "product": true. - A scene that carries straight on from the one before, same place, same moment, no cut, has "continues": true. It cannot add a character, change place or show the product for the first time; start a new scene for that. Use it when one action runs longer than 10 seconds. - Every scene is set in one of the places, by its id. Do NOT write an end card, logo card or pack shot on a plain background as a scene: the workflow adds the logo end card itself after the last scene. The product's hero moment happens inside one of the places. - A scene that is not continued starts from a picture made from the place's sheet and the characters' sheets: at most 5 characters in it with the product, 6 without. First write the storyboard for a person to read: the formula, then one block per scene (SCENE n, seconds, place, who, what happens, the narration, whether it carries on). Then the same ad as a JSON block, exactly this shape, in a fence that opens with three backticks and the word json: {"aspect":"9:16","mood":"the animation look for every picture and clip in one line: style, light, colours, texture","tagline":"the end card words or empty","narrator":"the narrator's voice: who, accent, age, tone", "characters":[{"id":"short-id","name":"Name","look":"a precise design: shape, size, colours, face, outfit, one detail","voice":"","faces":"the expressions this character needs, e.g. smug, worried, panicking, beaten"}], "places":[{"id":"short-id","look":"a precise design of the place: layout, furniture, light, colours"}], "scenes":[{"seconds":6,"place":"place-id","characters":["character-id"],"product":false,"continues":false,"action":"what we see happen, in one or two sentences, with the faces","narration":"the narrator's words over this scene, or empty"}]} Nothing after the JSON block. Brand pack: {{brand}} Brief: {{in}}
- 3Build the ad's stepsBy hand
Free. Writes a sheet per character and place, a picture per scene, one clip per scene, the narration and the join below. Build again after changing the storyboard.
Animated ad plan 10-09: formulas from the 40 X breakdowns and GetHookd ads, steps from Simon - Product spot
- 1BriefBy hand
Product: (name and one line on what it is) Tagline: (the line on the end card, e.g. "Comfort in one call"; blank = logo only) The idea: (one or two lines, e.g. "the bike at night on an empty track, speed and control") Tone: (e.g. cinematic and moody, warm and family, bold and fast) Story beats: (three to six beats in order, the last one the product's hero moment) Person: (who is in it, a few words on their look, or "no person") Where it runs: (TV, YouTube, website hero, social; this sets the shape)
- 2Treatment and shot listClaude Opus
You are the director of a 30-second brand commercial. Write the treatment from the brief and the brand pack. Plain labelled lines, in this order: IDEA: one sentence. LOOK: the grade, the light, the lens, the texture, in plain specific words. The brand's colours by their hex codes from the brand pack, and where each one shows (the light, the set, the wardrobe). WORLD: the one place the ad lives in, described for a set designer, no people. PERSON: the person from the brief in a few words (face, hair, build, outfit), or "none". Then exactly three clips, about 10 seconds each, each one two or three shots that cut inside the clip. For each: CLIP n: its beat in one line. SHOT 1 (00-03s): framing (extreme wide / wide / medium / close-up / macro / ground-level / overhead), the action, the camera move. SHOT 2 (03-06s): the same. SHOT 3 (06-10s): the same. SOUND: what is heard (engine, wind, footsteps, a pour). Never music, never speech unless the brief asks for a line. The product appears in every clip. Clip 3 ends on the product's hero shot, held still for the last two seconds. Vary the framings across the ad, never two wides in a row. Use only facts in the brief and the brand pack. Nothing else in your answer. Brand pack: {{brand}} Brief: {{in}}
- 3Product sheetNano Banana Pro
A product reference sheet of the product from picture 1, on one plain light-grey studio background, even soft light: the product seen from the front, the back, the left side, three-quarter front, from above, and one close-up of its label. Six views in a neat grid with thin gaps. The product exactly as in picture 1 in every view, its shape, colours, label and every printed word unchanged. Hyper-real product photography, crisp. No added text, numbers, labels or logos.
- 4World sheet wordsClaude Sonnet
You write the picture prompt for a director's framing reference contact sheet of the commercial's world: ONE image holding six numbered panels in a 3x2 grid with thin black gaps, all in the same grade, light and colours. Use the treatment's LOOK and WORLD lines and the hex colours word for word. The six panels: 1 extreme wide establishing shot, 2 wide lateral, 3 medium, 4 close-up of a surface or detail, 5 ground-level low angle, 6 a Dutch angle. Panels 1, 3 and 5 empty of people; panels 2, 4 and 6 may hold the treatment's PERSON if there is one. No product, no logos, no text in the pictures, only the small panel numbers. Cinematic film still, the grade named plainly (e.g. teal shadows, warm amber highlights, soft haze). Under 140 words. Nothing else in your answer. Treatment: {{in}}
- 5World sheetNano Banana Pro
Six framings of the ad's world in one grade: the look every clip copies.
- 6Person sheet wordsClaude Sonnet
You write the picture prompt for a character turnaround sheet of the treatment's PERSON: one image, plain mid-grey backdrop, even soft light, four full-body views (front, three-quarter, side profile, back) and three head-and-shoulders portraits (front, three-quarter, profile). The same person, outfit and hair in every view. Real skin texture, no beauty filter. No text, labels or numbers. Describe the person only from the PERSON line, and their outfit in the treatment's colours by hex code. If PERSON is "none", answer only: "No person in this ad, skip this step." Under 120 words. Nothing else in your answer. Treatment: {{in}}
- 7Person sheetNano Banana Pro
Only if the ad has a person. Then choose it as the third picture in each clip.
- 8Clip 1 wordsClaude Sonnet
You write ONE video prompt for CLIP 1 of the treatment below, about 10 seconds, its two or three shots cutting inside the one clip. The video model gets reference pictures: picture 1 is the product sheet, picture 2 is the world sheet, picture 3 (if there is one) is the person sheet. Refer to them as @image1, @image2, @image3 and never describe how the product or person looks: the pictures are the product and the person. The world, grade and colours come from @image2. Write exactly these parts, labelled: SCENE AND MOOD: one line, with the treatment's LOOK and hex colours. FRAME MAP: each shot of CLIP 1 with its seconds, e.g. "SHOT 1 (00-03s): ...". For each: START FRAME (what we see first), MOTION (one action, one camera move), END STATE (how the shot ends). Name the cut between shots: "CUT TO". LOCKS: the product exactly as @image1, its label and every printed word unchanged and readable; the person's face, hair and outfit as @image3; the grade and colours as @image2. Name what must not move (e.g. "the product rigid, only the light moves"). SOUND: the treatment's SOUND line. ABSOLUTELY NO MUSIC. No speech unless the treatment gives a line. AVOID: morphing, melting, the label changing, extra products, text or captions appearing, slow motion unless asked. Under 220 words. Nothing else in your answer. Treatment: {{in}}
- 9Clip 1Seedance 2.5
Make two or three takes and pick the best. 480p to draft (about $1.40 a take); set 720p for the keeper. Picture 1 the product sheet, picture 2 the world sheet, picture 3 the person sheet if there is one.
- 10Clip 2 wordsClaude Sonnet
You write ONE video prompt for CLIP 2 of the treatment below, about 10 seconds, its two or three shots cutting inside the one clip. The video model gets reference pictures: picture 1 is the product sheet, picture 2 is the world sheet, picture 3 (if there is one) is the person sheet. Refer to them as @image1, @image2, @image3 and never describe how the product or person looks: the pictures are the product and the person. The world, grade and colours come from @image2. Write exactly these parts, labelled: SCENE AND MOOD: one line, with the treatment's LOOK and hex colours. FRAME MAP: each shot of CLIP 2 with its seconds, e.g. "SHOT 1 (00-03s): ...". For each: START FRAME (what we see first), MOTION (one action, one camera move), END STATE (how the shot ends). Name the cut between shots: "CUT TO". LOCKS: the product exactly as @image1, its label and every printed word unchanged and readable; the person's face, hair and outfit as @image3; the grade and colours as @image2. Name what must not move (e.g. "the product rigid, only the light moves"). SOUND: the treatment's SOUND line. ABSOLUTELY NO MUSIC. No speech unless the treatment gives a line. AVOID: morphing, melting, the label changing, extra products, text or captions appearing, slow motion unless asked. Under 220 words. Nothing else in your answer. Treatment: {{in}}
- 11Clip 2Seedance 2.5
Two or three takes, pick one. Same pictures as clip 1.
- 12Clip 3 wordsClaude Sonnet
You write ONE video prompt for CLIP 3 of the treatment below, about 10 seconds, its two or three shots cutting inside the one clip. The video model gets reference pictures: picture 1 is the product sheet, picture 2 is the world sheet, picture 3 (if there is one) is the person sheet. Refer to them as @image1, @image2, @image3 and never describe how the product or person looks: the pictures are the product and the person. The world, grade and colours come from @image2. Write exactly these parts, labelled: SCENE AND MOOD: one line, with the treatment's LOOK and hex colours. FRAME MAP: each shot of CLIP 3 with its seconds, e.g. "SHOT 1 (00-03s): ...". For each: START FRAME (what we see first), MOTION (one action, one camera move), END STATE (how the shot ends). Name the cut between shots: "CUT TO". LOCKS: the product exactly as @image1, its label and every printed word unchanged and readable; the person's face, hair and outfit as @image3; the grade and colours as @image2. Name what must not move (e.g. "the product rigid, only the light moves"). SOUND: the treatment's SOUND line. ABSOLUTELY NO MUSIC. No speech unless the treatment gives a line. AVOID: morphing, melting, the label changing, extra products, text or captions appearing, slow motion unless asked. The last two seconds hold still on the product's hero shot. Under 220 words. Nothing else in your answer. Treatment: {{in}}
- 13Clip 3Seedance 2.5
Two or three takes, pick one. It ends on the hero shot of the product.
- 14Last moment of clip 3Pick a frame from a video
Free. The end card move starts here.
- 15Logo end cardNano Banana Pro
An end card: the logo from picture 1, centred, exactly as given, never redrawn, on a clean background in the brand's main colour with a soft glow behind the logo and a subtle film grain. If the brief below has a tagline, print it small under the logo, spelled exactly as written in quotes; if it is blank, no words. Nothing else in the picture. Brief: {{in}}
- 16Logo end moveKling 3.0
MOVE: from the first frame, a short light flicker and a quick glitch transition, then the logo end card resolves and settles. TIMING: 0-1s the first frame, 1-2.5s flicker and glitch, 2.5-4s the logo card resolves, 4-5s hold still on the logo card. CAMERA: locked off. KEEP FIXED: the logo and the tagline exactly as in the last frame, never changing shape or letters. AVOID: morphing letters, extra text, a different logo, shaking at the end.
- 17The commercialJoin the clips over music
Free. Upload the music track (Suno or a licensed track; the clips have none). Type the seconds kept from each clip, e.g. 6, 5, 7, 4, to cut each one to its best part.
Dreamina harvest: Honda, Nocturne, Lord Kelvin boards - Flyer · real parts
- 1BriefBy hand
A handout flyer for the brand in the brand pack. The picture step makes only the background; the real logo, products, QR code and words are laid on top afterwards. Mood: bold, premium, a night-time feel that makes the products look worth having.
- 2Background wordsClaude Opus
You write the picture prompt for the BACKGROUND of one printed flyer, portrait 2:3. The brand's real logo, product photos, headline, offer and QR code are laid on top of it afterwards by another step, so the picture must hold none of them. Picture 1, if given, is the STYLE reference: tell the model to make the background in the style of picture 1 (its lighting, colour, texture, era and mood). Never copy its words, people, objects or any watermark. The background must have: no text, no letters, no numbers, no logos, no products, no packaging, no people, no QR code, no frames or panels. A calm, open stage across the middle of the picture where products will stand (a surface, a floor, a glow or a horizon they can sit on). A simpler, darker top where a logo and a headline will sit. The bottom third can be plain: it will be covered. Use the brand pack's colours as the accent colours, by their hex codes. Name the light and the lens in plain specific words (for example: low magenta rim light, 35mm, shallow haze). No "masterpiece", no "8k", no "ultra detailed". Under 120 words. Use only facts in the brief and the brand pack. Nothing else in your answer. Brand pack: {{brand}} Brief: {{in}}
- 3BackgroundGPT Image 2.5
- 4FlyerLay on the real logo, products, words and QR
MotionViz method (x.com/Motion_Viz) - Flyer · style plate
- 1Style plateGPT Image 2.5
The front of a printed direct-mail leaflet, one full page, portrait. Make it in the style of picture 1 and follow its layout (its composition, era, print texture, type feel and ornament), with the logo, the products, the words and the QR code taking the places of picture 1's own elements. Keep picture 1's own lighting, ground, texture and mood; the brand colours ({{colours}}) become the accent colours inside that style. Never copy picture 1's words, objects or any watermark. Leave the places for the logo, the products, the words and the QR code styled but empty: no logo, no products, no packs, no words, no letters, no QR code. They are added afterwards. Super high quality, print-ready, crisp.
- 2FlyerLay on the real logo, products, words and QR
The approved flyer prompt (09-29), products and words laid on - Flyer
- 1BriefBy hand
- 2Flyer wordsClaude Sonnet
You write the picture prompt for one printed flyer. The brand pack below has the brand's colours and the products on this flyer; the brief has the ask, the offer and the words. The image model is given pictures, in this order. Refer to them by number and never describe what is in them: - Picture 1 is the STYLE and LAYOUT reference: tell the model to make the flyer in the style of picture 1 and to follow its layout (its composition, where its big and small things sit, its era, print texture, type feel and ornament), with the logo, products, words and QR code taking the places of picture 1's own elements. Never copy its words, its objects or any watermark in it. - Picture 2 is the brand's logo: show it clearly and exactly as given. - The next pictures are the products, in the order listed in the brand pack: place each one exactly as given, real and sharp, never redrawn. - The last picture is the QR code: show it exactly as given, flat, square and uncropped, on a plain light panel, with the call to action printed right beside it. Write one prompt: the headline, the offer, the call to action and every other word printed on the flyer, exactly as it must appear and spelled out in quotes. Do not decide where things go: never write top, bottom, centred, row, column or grid; say the layout follows picture 1. Keep picture 1's own lighting, ground (dark or light), glow, texture and mood; the brand pack's colours become the accent colours inside that style (in a neon style they are the glowing colours), never a reason to lighten or flatten it. Ignore any photography or lighting rule in the brand pack (natural light, studio light, editorial product photography, clean retail feel): those are for product photos, not flyers. Leave the model creative freedom on the rest. Name only the products listed. Use only facts in the brief and the brand pack. Nothing else in your answer. Brand pack: {{brand}} Brief: {{in}}
- 3FlyerGPT Image 2.5
Mainframe creative expertNot ready for social - 3D product film
- 1The briefBy hand
Product: BackRelease Pro, a curved back device you lie on, with five forces working together for back relief Parts: from the top down. The white outer shell = "EMS Pulse Activation" (the two silver plates on the black pad under it). The row of small light dots on the black pad = "Red/Blue Light Therapy". The warm heat layer under the pad = "Deep Heat Therapy". The vibration motor = "Neuro-Vibration". The curved base = "Dynamic Spinal Traction" The film: fast and dynamic. It spins around to show the device, the camera moves up close, then pulls back a little further away. Then it opens quickly: the top lifts and its name pops on, the next section comes out and its name pops on, and so on for each part. Then it goes back together, spins around and rests. Then it lifts up to show the decompression (spinal traction) and comes back down to rest. End card: BackRelease Pro · Five Force Technology Look: clean light-grey studio, soft shadows, a gentle light sweep across the surfaces Shape: 16:9
- 2Shot listClaude Opus
You are the director of a short hyper-real 3D product film. Turn the brief into a shot list. Each shot is ONE camera move between two still pictures: the picture it starts on and the picture it ends on. The video model films the move between them, so what changes between the two pictures (the framing and the product's state) IS the shot. Rules from our course material: - One named camera move a shot (orbit left 180 degrees, fast push in to a close-up, pull back to a wide, crane up, locked off). Never two moves blended in one shot: a spin is one shot, a pull back that opens the parts is another. - Contrast keeps it alive: cut wide to close-up, fast to still. A fast shot is 3 seconds; a hold or a big reveal is 4 to 5. - When parts open one by one, each part is its own shot (3 s) that ends on that part out and gives its name. Open parts stay open in the next shot unless it closes them. - When the product goes back to an earlier state and framing, end the shot on that earlier picture (end_same_as) so it matches exactly. - A gentle specular light sweep across a surface adds sparkle; use it in a hero or a resting shot, not every shot. - No words on screen except the names (the Bench lays those on) and the end card. - Follow the brief's order and pace. If the brief is one line, make one shot. - Use only parts and names given in the brief. First the shot list in plain lines: shot number, seconds, camera, what happens, the picture it ends on, the name if any. Then ONE JSON block: ```json {"aspect": "16:9", "look": "background, light, lens, grade for the whole film", "tagline": "the brief's end card words, or empty", "start": "the first picture: the framing and the product's state", "shots": [{"seconds": 3, "camera": "fast orbit 180 degrees to the left", "action": "what moves, in one or two sentences", "end": "the picture it ends on: framing and the product's state", "end_same_as": null, "label": "Name = the part it points at (e.g. Deep Heat Therapy = the warm heat layer under the pad), or empty"}]} ``` end_same_as: null for a new picture; 0 to end on the start picture; N to end on the picture shot N ended on (then "end" may be empty). 1 to 12 shots, 3 to 10 seconds each, 60 seconds at most in all. Brief: {{in}}
- 3Build the film's stepsBy hand
Free. Writes one picture per new product state or framing, one shot between each pair, the names, the end card and the join below. Build again after changing the shot list.
Simon's 10-08 brief + Seedance bible 535-552, Kling 3 guide, Seedance 2.5 running logNot ready for social - Product in motion (3D)
- 1BriefBy hand
Result: (one of: exploded view / how it works / 360 spin / reveal / logo) Product: (its name and one line on what it is) Parts or mechanism: (exploded view: the parts inside, from the outside in. How it works: what happens inside, e.g. "the lipid shell wraps the peptide and keeps it safe until it reaches the gut". Leave empty for spin, reveal or logo) The move: (one line, from start to end, e.g. "the device slowly comes apart, each layer floating out") Background: (e.g. plain soft grey studio, or dark with a soft glow behind) Where it goes: (website hero, product page, ad)
- 2Start picture wordsClaude Opus
You write the picture prompt for the FIRST frame of a short product animation. The image model gets picture 1: the real product photo. Never describe the product's shape, colours, label or words; picture 1 is the product. Say "the product from picture 1, exactly as it is, its label and every printed word unchanged". Write the first frame for the brief's Result: - exploded view, 360 spin: the product whole, upright, centred, facing the camera, on the brief's background. - how it works: the product whole in the same way. If the brief's mechanism happens inside, show the product whole here; the inside is shown at the end. - reveal: the product hidden in shadow or under a soft cover, only an edge of light on it. - logo: the background only, empty, ready for the logo to appear. Name the light and lens plainly (e.g. soft key light from the upper left, a gentle rim light, 85mm, shallow depth). Hyper-realistic 3D product render, clean, crisp, premium. Leave space around the product for movement. No text, no captions, no added logos. Under 100 words. Nothing else in your answer. Brief: {{in}}
- 3Start pictureNano Banana Pro
Picture 1 is the real product photo. Check it before going on: the label and every word must be exact.
- 4End picture wordsClaude Opus
You write the picture prompt for the LAST frame of a short product animation. The image model gets picture 1, the first frame, and picture 2, the real product photo. Keep picture 1's background, light, lens and framing exactly; change only what the brief's Result changes. Never redraw the product's label or words; they stay exactly as in picture 2. Write the last frame for the brief's Result: - exploded view: the same product taken apart along one axis, every part from the brief's list floating in order with even gaps, outside parts furthest out, the centre part still in place, all parts in the product's own materials. - how it works: a hyper-realistic 3D cutaway of the brief's mechanism (e.g. a cross-section showing the shell around the core), clear and scientific, in the product's colours. - 360 spin: the same product seen from the back. - reveal: the same product fully lit, the hero shot. - logo: the brand logo from picture 2 centred on picture 1's background. Say what changes in one or two plain sentences, then "everything else exactly as in picture 1". No text, labels or captions added. Under 100 words. Nothing else in your answer. Brief: {{in}}
- 5End pictureNano Banana Pro
Made from the start picture, so both sit in the same scene. Check both pictures before paying for the video.
- 6Move wordsClaude Sonnet
You write ONE video prompt for a 5-second product animation. The video model gets the first frame and the last frame, so it already knows how both look: write MOTION ONLY, never describe the product. Use the brief's Result and The move. Write exactly these parts, labelled: MOVE: what happens between the first and the last frame, in one sentence. TIMING: 0-1s hold on the first frame, 1-4s the move, 4-5s settle and hold on the last frame. (Logo: 0-2s the logo appears, 2-4s it settles, 4-5s hold.) CAMERA: one slow move only (a slow push in, or a slow orbit for a spin), or locked off. KEEP FIXED: the background, the light, the label and every printed word, the parts' shapes and materials. AVOID: morphing, melting, parts changing shape, extra parts, text appearing, camera cuts, shaking. Smooth, premium, slow. Under 90 words. Nothing else in your answer. Brief: {{in}}
- 7The moveKling 3.0
Kling films the move from the start picture to the end picture. 720p to test; set Quality to 1080p for a keeper.
- 8Label the partsLabel the parts (real words, a line to each part)
Free. For an exploded view or how it works: write one label a line, Name = what it points at. Claude finds each part in the last frame; the Bench holds that frame and fades each name in with a line to its part, in a real font, so no word is ever drawn by the video model. Also gives the labelled still for a product page. Leave the labels empty and skip this step for a spin, reveal or logo.
- 9UpscaleTopaz video upscale
Only on the take you keep. Takes the labelled video; point it at @move instead when there are no labels.
Cookbook: start and end frame workflows + our 3D formulas - Promo film
- 1The briefBy hand
The idea: the chaos of the peptide internet against one calm woman's morning. VERO is the simple, done-for-you answer. Characters: a gym bro in his late 20s, tank top, filming himself live on TikTok, loud and fast. A girl influencer in her mid 20s, ring light, pastel room, earnest and over-technical. Jane, early 40s, calm, fit, natural, in her bright kitchen. Places: the gym bro's bedroom with a mini fridge; the influencer's pastel room with a whiteboard; Jane's bright kitchen; Jane's hallway. The story: 1 the gym bro live on TikTok showing a stack of twenty peptide vials and needles on his desk. 2 cut to Jane waking up, calm morning light. 3 the influencer with a dosing calculator on her phone and a whiteboard full of molecule drawings, explaining. 4 cut to Jane taking the VERO jar from the cupboard and taking her dose. 5 the gym bro at a fridge of twenty-five vials, prepping needles. 6 cut to Jane putting the jar away and picking up her tennis bag, out the door. Lines: the gym bro: "Day forty of the stack, twenty peptides, let's go." The influencer: "So you reconstitute with two mils, then it's ten units on the syringe..." Jane says nothing. Length and shape: about 60 seconds, 16:9. End card line: Peptides, made simple.
- 2StoryboardClaude Opus
You are the director and storyboard artist of a short promo film. Lay out the film from the brief and the brand pack. Rules of the craft: - Every scene is one clear moment a viewer reads with the sound off: who, where, doing what, in one sentence. - Every person is seen: never a hand alone, a screen alone or a faceless shot as a whole scene. - When storylines cut between each other, each scene still stands on its own, and the contrast is plain. - The product is shown so a viewer knows what it is and how it is used, using only facts in the brand pack and the brief. It appears in a scene only where the story puts it. - Lines are said word for word as the brief gives them, at most about 2.5 words a second of the scene. Add no lines the brief does not ask for. - A narrator tells the story over the film (an outside voice, never a person on screen): in each scene with no spoken line, write the narration, at most 2.5 words a second. Read it aloud in your head: it must make sense with the pictures off. - Hands (Simon 10-15, the tablets scene): never pour, tip, empty or catch small things (tablets, capsules, powder, drops) into a hand or a glass, and never a reach, grab or lift; start on the moment just after (the tablets already resting in the open palm, the glass already full). - Scenes are 4 to 12 seconds, 3 to 12 scenes, the whole film at most 90 seconds. At most 4 characters and 4 places in the film. In one scene: at most 2 people when the product is in it, 3 when it is not. First write the storyboard for a person to read: one block per scene (SCENE n, seconds, place, who, what happens, the line if any). Then the same film as a JSON block, exactly this shape, in a fence that opens with three backticks and the word json: {"aspect":"16:9","mood":"the look of the whole film in one line: grade, light, lens, texture","tagline":"the end card words or empty","narrator":"the narrator's voice: accent, age, tone", "characters":[{"id":"short-id","name":"Name","look":"age, face, hair, build, outfit, one detail","voice":"accent, pitch, pace, energy"}], "places":[{"id":"short-id","look":"the place for a set designer, its light, no people"}], "scenes":[{"seconds":8,"place":"place-id","characters":["character-id"],"product":false,"action":"what we see, in one or two sentences","line":{"who":"character-id","words":"the words"},"narration":"the narrator's words over this scene, or empty"}]} Use "line": null for a scene with no line; a scene has a line or narration, never both. Nothing after the JSON block. Brand pack: {{brand}} Brief: {{in}}
- 3Build the film's stepsBy hand
Free. Writes a sheet per character and place and one clip per scene below. Build again after changing the storyboard.
Promo film plan 10-04: storyboard first (Dreamina story films, Seedance bible)Not ready for social - Street interview · short prompt
- Not run1Script
- Not run2Start frame
- Not run3Clip plan
- Not run4Video prompts
- Not run5Video clips · Omni Flash
- Not run6Check
- Not run7Final video
application.md - UGC ad · from a real creator
- 1Creator video + briefBy hand
Product: (name, what it does, the one fact to say, its real size, e.g. a small jar that fits in one hand; it is in their hand from the first moment, never picked up) New person: (age, gender, a few words on their look) Setting: (where they film it, e.g. their kitchen in the morning) What they say (any length: a short line makes one clip, a longer script is cut into several clips joined together): "..." Tone: (e.g. surprised, matter-of-fact, excited but quiet)
- 2Creator framePick a frame from a video
One full-size frame of the creator's video. Change the second to pick a better moment.
- 3Creator voiceTake the real voice (from the creator video or a voice clip)
The real voice sample, cut from the creator video with its room left in (car, studio, street): Seedance speaks the script in this voice, copying how it talks, as the Ads dashboard does (Simon's rule: every real UGC voice comes from a real sample, never invented). Move 'From second' to a stretch where they talk with no music. To use a different real voice, upload its clip here instead.
- 4Creator movementCut the creator's movement (no sound)
A silent piece of the creator video. Seedance copies how the person moves, talks with their hands and handles things, never their face, body or room. Move 'From second' to a lively stretch.
- 5Creator fileClaude Opus
You write the creator file for one AI UGC creator, from the brief below. The creator is a NEW, believable everyday person, never a model and never a real person. Write it in plain labelled lines, in this order: IDENTITY LOCK: age, gender, face shape, skin, hair (colour, length, how it falls), eyes, build, and 3 signature details that make the face recognisable (e.g. a small scar over the left eyebrow, a gold stud in one ear, a slightly crooked front tooth). Add ONE deliberate flaw that repeats in every shot (e.g. flyaway hair at the crown, a chipped nail). OUTFIT: one everyday outfit with colours and one accessory. VOICE: never invented. The voice is the real voice sample cut from the creator video (or a real voice clip), sent to the video model as it is. Note only how they open a sentence and one filler word they use, so the words fit that voice. CAMERA HABITS: front phone camera, 24-28mm feel, held at arm's length, no beauty filter, eye level or slightly below, small natural wobble. NEVER CHANGE: the face, hair, signature details, flaw, outfit, accessory. Then a line "SHOT:" and copy the brief's product, setting, spoken line and tone word for word. Use only the brief. No commentary. Brief: {{in}}
- 6Casting cardGPT Image 2.5
A casting reference card of one person, a 2x2 grid of four photos on a plain mid-grey studio backdrop, even soft light: top left facing the camera, top right three-quarter view, bottom left side profile, bottom right from behind. The same person, same outfit and same hair in all four, full head and shoulders to the waist. Real skin texture, natural, no retouching, no beauty filter. No text, no labels, no numbers, no borders between the photos other than a thin gap. The person, from this creator file (use only the IDENTITY LOCK and OUTFIT lines): {{in}}
- 7Still promptClaude Sonnet
You write the prompt for ONE vertical start frame of a UGC ad: a still from a phone video the person films of themselves. The image model gets three pictures. Refer to them by number and never describe what is in them: - Picture 1 is a frame from a real creator's video: use it ONLY for the phone-camera feel, the light and the kind of room. Never copy its person, face, clothes, words or any logo in it. - Picture 2 is the casting card: this is the person. Keep their face, hair, signature details and outfit exactly as in picture 2. - Picture 3 is the product: show it exactly as given, label turned roughly to the camera, already held loosely in one hand at the first moment, at its real size from the SHOT lines (say that size in the prompt). Exactly one of it in the entire scene: no second one on any shelf, counter or surface, no copy in a mirror or reflection. Write only the placement, the scene and the action: where the person is (the SHOT setting), the camera angle (front camera at arm's length, eye level or a little below). The picture IS what the phone's camera sees, so the phone itself never appears in it: no phone in hand, no phone in a mirror, no phone floating at the edge, what their hands and face are doing at the first moment of the spoken line, the light named plainly (e.g. soft window light from the left). Phone photo quality, real skin, no beauty filter, slight imperfection. No text, captions or words drawn anywhere in the picture. Under 110 words. Nothing else in your answer. Creator file: {{in}}
- 8Start stillGPT Image 2.5
Picture 1 the creator frame (camera feel only), picture 2 the casting card (the person), picture 3 the product from the brand.
- 9Cut the script into clipsBy hand
Free. Cuts the spoken words in the brief into clips (about 3 words a second, at most 30 words a clip, as in the Ads dashboard): a short line makes one clip, a long script four, five or six. For each piece: a prompt, then the clip, where Seedance speaks the words in the real voice clip's voice (the Ads dashboard method), moving like the creator. Then the join and the phone finish.
Simon's X saves, UGC upgrades U1-U4Not ready for social