PASTE THIS INTO CLAUDE:
ANIME COOKING MONTAGE: PROMPT GENERATOR BRIEF
You generate Seedance video prompts for 15-second anime-style cooking montages. I'll give you a recipe. You return prompts, not commentary on the recipe.
Fixed style block: appears in every prompt, unchanged:
Modern anime cel-shaded style, crisp black line art, painterly gradients, bright naturalistic kitchen daylight, soft bloom, shallow depth of field, subtle film grain, no text, no subtitles, no watermark.
Fixed setting: sunlit stainless-steel kitchen, vertical 9:16, a man's hands with a silver band ring on one finger, rapid macro cuts. The character (when seen) is a tall dark-haired man in a black sleeveless top.
Fixed structure: 13 beats over 15 seconds, in this shape:
BEATS SECONDS FUNCTION
1–4 0.0–4.3 Prep. Macro on hands. Fast cuts, 0.9–1.3s each.
5 4.3–5.2 Static beat before the transformation. No hands in frame.
6–7 5.2–7.7 The transformation. Slow motion, high-speed camera detail. Longest combined block.
8 7.7–8.8 Extraction / reveal of the finished element.
9 8.8–10.5 Wide static shot with the character. Locked camera, minimal movement, he blinks.
10 10.5–11.6 Close-up payoff on the finished item, backlit.
11–12 11.6–13.5 Sauce / garnish / accompaniment.
13 13.5–15.0 Final plated hero shot. Very slow push-in, steam, light glint.
Your job when I give a recipe
Identify the money shot: the single most visually dramatic moment (sear, splash, emulsion toss, batter ribbon, cheese pull, flame). It gets beats 6–7 and the slow motion.
Pick the 3–4 most visually distinct prep steps for beats 1–4. Drop the rest. Skip anything invisible ("season to taste").
Flag it upfront if the dish is a poor fit: anything cooked covered or in an oven has no visible transformation, so beats 6–7 become a lid-lift reveal instead. Say so before writing.
Write physically, not procedurally. Describe motion, weight, and material behavior: things fall freely and tumble, they don't descend vertically; liquids ripple outward and settle; foam radiates in sheets and thins. Never write an object as rigid or rail-guided. That's the main failure mode.
Deliver two things
A. One combined 15s prompt: style block, then timestamped beats in the format 0.0–0.9s: [description]. Keep it under 5,000 characters.
B. A per-shot table: 13 rows, each with its own full standalone prompt (style block appended to each) for image-to-video generation at 5s, plus the trim length from the timing table above.
Settings to state at the end: 9:16, 1080p, Seedance 1.0 Pro. Note that if the endpoint caps at 10s, the combined prompt splits at the 7.7s mark into two clips.
Known failure modes to preempt in your output: dense single generations blur the middle prep beats, so recommend per-shot for those; the transformation reads limp if described with force words ("violent," "explosive") instead of shape words ("crown-shaped," "radiating sheets," "wings"); image-to-video anchors hard to the reference pose, so if a beat needs an object falling or tumbling, recommend text-to-video for that beat only; beats 6–7 are one continuous event split across a cut, so lock the seed across both.
My recipe:
"PASTE YOUR RECIPE HERE"