Character Fusion Studio · skill pública Open Skill

Character Fusion Studio

Character Fusion Studio é uma Open Skill pública no Digen. Abra para ver o que este fluxo cria.

@ragnaesti

Anime

3 0 4

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4167013/2026-09-28/aw4ym5ff.png

# Character Fusion Studio

## When to use
- The user wants to fuse/combine two characters (from uploaded images or text descriptions) into one new character.
- The user wants a full-body image of the fused character plus a short vertical transformation video.
- The user wants to redo/restyle the fused character, regenerate one phase clip, or re-mix the video.

## Reference files
- `references/fusion-workflow.md` — Read this before starting. Contains the camera-language and color/film-emulation guidance, per-phase video prompt templates, and the exact deliverables format.

## Steps
1. **Gather the two references.** If the user provided two images, run `recognize_image` on each to extract face, silhouette, outfit, colors, powers, and personality. If they only gave descriptions, use those directly. If references are missing, ask for them before proceeding.
2. **Read `references/fusion-workflow.md`** for the camera-language and color-grading guidance and the prompt templates.
3. **Design the fused character.** Combine the two sources into one cohesive character: preserve visible traits from both, give it a distinct face, readable silhouette, and consistent costume. Note which trait came from which source. Pick a name.
4. **Choose camera language and color grade** per the reference file, matching the user's requested style (or pick a suitable style and briefly explain the choice).
5. **Set the background.** Ask the user whether they want to provide a background image reference or have one generated. If they provide an image, use it as a reference. If not, generate a suitable background with `generate_image` (`model='t2i'`, `orientation='portrait'`, `name=<fused name> background`, `entity_type='locations'`) matching the chosen camera language and color grade. Use this background consistently across the fused character image and all video clips.
6. **Generate the full-body fused character image** with `generate_image`:
   - If two reference images exist: `model='i2i'`, `input_urls=[ref1, ref2]`, `orientation='portrait'`, `name=<fused name>`, `entity_type='characters'`.
   - If only descriptions: `model='t2i'`, `orientation='portrait'`, `name=<fused name>`, `entity_type='characters'`.
   - Incorporate the chosen background into the prompt (or include it in `input_urls` if the user provided one) so the scene matches.
   - The prompt must bake in the chosen camera language and color treatment so the video matches the image.
   - If the fused image shows duplicate limbs, split faces, or inconsistent costume, regenerate it until the fusion is clean and cohesive — aim for the best possible fusion outcome.
7. **Generate the four phase clips** with `generate_video` (`model='ref2v'`, `orientation='portrait'`) to keep identities consistent throughout:
   - Intro & fusion clips: pass the two original character images as references (`input_urls=[ref1, ref2]`) so both characters stay consistent while they meet and fuse.
   - Reveal & signature clips: pass the fused character image as reference (`input_urls=[fused_image]`) so the fused identity stays consistent.
   - Use the per-phase templates in the reference file. Use `name` like "intro clip"/"fusion clip"/"reveal clip"/"signature clip", `entity_type='storyboard_videos'`.
   - If a clip shows duplicate limbs, split faces, or flickering costume, regenerate it until the fusion is clean.
8. **Mix the clips** with `mix_clips` in order (intro → fusion → reveal → signature), `orientation='portrait'`, `name=<fused name> final`, `entity_type='final_videos'`.
9. **Optional sound/music.** Generate a soundtrack with `generate_music` (`model='music'`, `name=<fused name> score`, `entity_type='music'`); optionally a voiceover with `generate_tts` (`model='tts.design'`, `voice_instruction` in English, `entity_type='voiceovers'`). Mix the audio into the final video if desired.
10. **Deliver** the fused character name + design summary, the image prompt, the Camera Language section, the Color/Film Emulation secti