Anime Soccer

anime-soccer

Turn the user into the hero of a 3-scene (~45s) Japanese-anime football short, but the star is the person running it. Each scene is a 15s cel-shaded anime beat with the user (named, in CAPS) and Japanese dialogue, generated on Seedance off an anime character sheet made from the user's photo, then stitched.

The proven pipeline (do not deviate from the order): user photo → anime CHARACTER SHEET (nano-banana-pro) → DESIGN the dramatic match + WRITE 3×15s scene prompts (you, acting as Claude) → SEEDANCE each scene off the sheet → STITCH.

The magic is in three things, in this order: (1) a character sheet so the user's anime likeness is consistent across every scene, (2) a specific, dramatic match designed for them (their team vs the opponent, real kit colors, a concrete goal mechanic each scene — not a generic "he scores"), (3) the scene-prompt formula (cel-shaded style tag + named hero in CAPS + camera/anime techniques + Japanese VO).

Stage 0 — Intake

Collect, in one pass (the only things that change per user):

• The user's photo (required) — a clear face photo; this becomes the anime hero. Local path or URL. Likeness comes from this. Upload local files via upload_asset → use the public_url. (If on Claude Desktop where pasted images don't reach MCP, ask once for a URL or a .zip.)
• On-screen hero name (optional) — what to call them in CAPS in the prompts + Japanese VO (e.g. "KAITO"). Default to their first name, or "THE STRIKER" if they'd rather not use a name.
• Their team (required) — the hero's side. Capture the kit colors / stripe pattern (e.g. "sky-blue and white vertical stripes"), NOT the crest or sponsor (describe colors, not trademarks — see de-branding in failure modes).
• The opponent (required) — who they're playing against. Capture the opponent's kit colors too. Opponent players (keeper, a defender) are generic — described by kit color, not named.

Confirm in one line ("Putting YOU in an anime match — [TEAM] vs [OPPONENT], you score the winner…"), then run the pipeline. No further yes/no gates.

Stage 1 — Anime character sheet (generate_image_edit, nano-banana-pro)

Generate ONE anime character sheet of the user from their photo. The sheet (multiple angles + full body + close-ups) is what keeps the face consistent across all three scenes.

Call generate_image_edit with:

• provider: `nano-banana-pro` (REQUIRED default). It does reference-identity → anime stylization cleanly for any real face. Don't use `gpt-image-2` here — OpenAI's safety system rejects anime-editing a real photo of an identifiable person (content_policy / image_generation_user_error).
• images: [<the user's photo url>]
• aspect_ratio: 16:9 · quality: high · output_format: png
• prompt (verbatim template — fill {NAME}; put them in their team kit): `` Create a character sheet of this football player named {NAME} in a {TEAM kit colors} kit, with close-ups and full body shots from different angles in Japanese anime style. ``

Save the returned URL as state.sheets[HERO]. Inspect: the sheet must be clearly that person, anime-styled, with usable full-body + face angles. Re-roll if it drifts. (If the user also wants a named teammate/rival on screen with a real face, make a sheet per character the same way; otherwise opponents stay generic.)

Stage 2 — Design the match + write the 3 scene prompts (you = "Claude")

This is the step the creator did in Claude. You do it directly. There's no real game to research — you design a dramatic, specific match for the user, then WRITE it in the formula. Specificity is what sells it: pick a real goal mechanic for each scene, not "he scores."

1. Design the match: the user's team ([TEAM]) vs [OPPONENT] in a huge night stadium. Decide the arc — the hero comes on with the team behind, equalizes, then scores a dramatic late winner. For each goal choose a concrete mechanic (a volley, a header off a cross, a turn-and-finish, a solo run, a free-kick) and a scoreline + minute. Use the two teams' kit colors throughout; opponent keeper/defenders by color only.

1. Write exactly 3 scenes, 15s each, following the proven 3-act arc:

• Scene 1 — "THE CALL": the hero comes on / kickoff for [TEAM], tension building, crowd roaring, a teammate or the bench urging them on.
• Scene 2 — "THE EQUALIZER": the hero scores the equalizer vs [OPPONENT] with the chosen mechanic (real assist build-up, the strike, the keeper beaten) — e.g. 1-1.
• Scene 3 — "THE WINNER": stoppage time, the hero scores the decisive goal, ending on the win + an on-screen final-score card.

1. Scene-prompt formula (match this structure exactly — it's what makes it work):

• Open with the style tag: Japanese anime action style, cel-shaded, high contrast, [color grade e.g. cold desaturated → blazing gold], [intensity], [setting: packed night stadium, the minute + scoreline].
• Beat-by-beat action naming the hero in CAPS with their kit colors, the exact goal mechanic, and anime camera/FX language: extreme close-up on the eyes, low heroic angle pushing up, slow dolly-in, anime smear frames, bullet-time on the strike, impact frame flashes pure white then SNAPS to full speed, net ripples, golden particle burst, motion-blur trail, frozen goalkeeper mid-dive.
• Mood line: e.g. brooding/determined → electric/triumphant → euphoric/explosive.
• End with Japanese VO/subtitle lines, one per speaker, prefixed by the character's name, plus a commentator — short dialogue + the hero's inner monologue. Format: `` VO/subtitle (JP): {NAME}「…」 / {NAME} (inner)「…」 / commentator「…」 ``
• Scene 3 also gets a final on-screen card: Final on-screen card: {TEAM} {SCORE}.

(Proven structural reference — the dramatic anime-match edit this format is modeled on; mirror its SHAPE exactly, swapping in YOUR hero name + YOUR team + the opponent: Scene 1 cold-graded close-up → strides on under floodlights, a teammate shouts, hero inner-monologue. Scene 2 a cross comes in → the HERO volleys it in, bullet-time + white impact flash, 1-1, keeper frozen. Scene 3 stoppage time, the HERO turn-and-spin finish past the defender, knee-slide, gold confetti, "[TEAM] 2-1" card. Each opens with the cel-shaded style tag and closes on JP dialogue + monologue + commentator. Reuse this shape; swap in your hero + teams + the goal mechanics you designed.)

Stage 3 — Generate each scene (generate_reference_video, seedance)

One call per scene. Use the character sheet as reference and map the hero name to the image via the Seedance @Image1 token.

Call generate_reference_video with:

• provider: `seedance` · seedance_backend: `ark`
• reference_images: the hero sheet (plus any extra character sheets used in THAT scene) — e.g. [sheets[HERO]]
• aspect_ratio: `16:9` · resolution: `1080p` · duration: `15` · sound: `true` (the JP dialogue + dramatic SFX are the point here — this format KEEPS generated voice/audio)
• prompt: a name→image mapping line, then the scene prompt. Example: `` {HERO} is @Image1. <full Scene 1 prompt from Stage 2> ` Reference the hero via its @ImageN` token where it first appears so Seedance binds the likeness.

Fire all three in parallel (background: true) and poll. Re-roll a scene if: the goal mechanic is wrong, the face drifts off the sheet, or the title/score card text garbles. If a scene trips OutputAudioSensitiveContentDetected, re-roll with a new seed (or, as a last resort, sound: false for that scene only and add the dialogue as an overlay).

Stage 4 — Stitch