Anime Fight Scene Generator · skill pública Open Skill
Anime Fight Scene Generator
anime fight scenes samurai finishes the demon
Anime
3 2 7
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4205557/2026-10-06/6ujnvpz8.jpg
# Anime Fight Scene Generator Turn a fight idea into animated anime fight videos. The skill decides the details itself: it designs the fighters and setting, generates anime-style fight stills, animates each key moment into a clip, and optionally stitches them into a short fight sequence with music. ## When to use - The user describes an anime fight (fighters, fighting style, setting, mood) and wants images, clips, or a full fight video. - The user wants a fight scene generated from a one-liner or idea, with no source images. - The user wants to continue/redo a fight scene from the same session. ## Steps 1. **Clarify only if truly needed.** If the user gives a clear fight idea, proceed on your own. If the request is ambiguous (e.g. just "make an anime fight"), ask one short question about the fighters/setting, then proceed. Default to portrait (9:16) for short-form; use landscape (16:9) if the user asks for cinematic. 2. **Plan the fight.** Break the idea into 2–4 key moments (e.g. clash, dodge, finisher). For each moment, decide: the two fighters, the setting, the action, and the anime style (shonen, dark fantasy, mecha, etc.). Use `write_todos` to list the steps and update as you progress. 3. **Generate anime fight stills** with `generate_image`: - `model='t2i'` (no reference image) or `model='i2i'` (if the user provided a character/reference image, pass it in `input_urls`). - `orientation='portrait'` by default, `'landscape'` if requested. - `resolution='720P'` (or `1080P` for higher quality). - `entity_type='anime_fight_stills'`, `name` = a short shot title (e.g. "Clash"). - Prompt in English: describe the two fighters (appearance, outfit, colors), the setting, the action/moment, dynamic motion lines, speed lines, impact effects, and the anime art style. Keep the same two fighters' look consistent across all stills by repeating their descriptions. 4. **Animate each still into a fight clip** with `generate_video`: - `model='i2v'` (image-to-video), passing the still's asset_id in `input_urls`. - Same `orientation` as the still. - `resolution='720P'`, `duration` ~4–5s. - `entity_type='fight_clips'`, `name` = the same shot title. - Prompt in English: describe the motion of that moment (e.g. "the swordsman lunges forward, blades clash with a burst of sparks, hair and coat whip in the wind, dynamic camera shake"). Keep motion consistent with the still. 5. **Optionally add a battle cry / impact voice** with `generate_tts` (`model='tts.design'`, `voice_instruction` describing an intense anime battle voice, e.g. "deep intense male anime voice, shouting a battle cry") — `entity_type='voiceovers'`. This is optional; skip if the user didn't ask for audio. 6. **Optionally generate background music** with `generate_music` — `entity_type='music'`, prompt describing an intense anime battle soundtrack (e.g. "epic orchestral anime battle theme, driving percussion, rising tension"). Optional. 7. **Stitch the clips into a fight sequence** with `mix_clips`: - `clips`: one entry per clip, each with `video.url` = the clip's asset_id, and optionally `audio.url` = the music asset (and/or the battle cry) to layer over it. - `orientation` matching the clips. - `name` = the fight title, `entity_type='final_fights'`. - If the user only wanted single clips, skip this step and deliver the clips directly. 8. **Reply** in natural language: summarize the fight you made, list the stills/clips/final sequence, and note that intermediate stills and clips are visible on the canvas. Call `set_guidance` to suggest next actions (e.g. "redo the finisher", "make it landscape", "add a different fighter", "extend the fight"). ## Notes - Keep the two fighters' designs consistent across all stills by repeating their visual descriptions verbatim in each prompt. - Report failures honestly; never fabricate URLs. If a generation fails, retry once with a slightly adjusted prompt. - Video generation does not support NSFW/ad