Character Studio · skill pública de Open Skill
Character Studio
Character Studio es una Open Skill pública en Digen. Ábrela para ver qué crea este flujo.
Anime
10 1 6
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4304237/2026-10-20/3fnn6atc.jpg
---
allowed-tools:
- generate_image
- generate_video
- generate_tts
- mix_clips
- caption_video
- probe_media
- extract_clip
- transcribe_audio
- create_document
---
# Talking & Walking Character Studio
Make short, charming videos where non-human characters (animals, birds, mammals, trees, vegetables, plants, objects, invented creatures) talk in character and physically walk around a scene.
## When to use
- The user wants a character that **speaks and walks** (e.g. "a talking carrot that walks to the market", "a chatty owl on a branch").
- The user asks for a cartoon/animated dialogue scene with one or more creatures.
- The user gives a species list ("animal, bird, mammal, tree, vegetable") and wants each one to talk.
- The user wants the character **in a real place (kitchen, street, market) or a made-up/stylized place (studio stage, fantasy world)**.
- The user wants the character to **dance, eat, cook, play, or otherwise behave like a person**.
- The user wants the character to **interact with objects/props** that are placed in the scene with them.
## Core idea
Audio drives the performance. Each speaking character gets:
1. a still character image (their look),
2. a voice track (their voice),
3. a video generated from **that image + that audio** (so the mouth movement matches the line and the body walks while speaking).
For richer scenes (real/stylized environment, props, dancing, eating), use **reference-to-video (`ref2v`)** instead: it takes the character image **plus** environment/prop reference images **plus** the voice audio, so the character stays consistent, stands in the right world, and physically handles the objects.
## Steps
1. **Plan first.** Use `write_todos` to list the scenes. Decide: the cast (species + personality), the **environment** (real or artificial), the **props/objects** the characters will interact with, the **action** per shot (walk / talk / dance / eat / play), the line for each character, and the scene order. If the user gave no dialogue, write short, funny, in-character lines yourself (1–2 sentences each, under ~10 seconds of speech).
2. **Read the character bible** for species design cues, voice profiles, environments and prop handling: `references/character-design.md`.
3. **Character image** — one full-body image per speaking character, so the body is visible (walking, dancing) and the face is visible (talking).
- `generate_image(model='t2i', prompt=..., orientation='portrait', resolution='1080P', name='<Character Name>', entity_type='characters')`
- Prompt must include: species, clothing/accessory, personality, art style, **full body visible head to toe, standing/walking pose, front three-quarter view, clean simple background**.
- Keep the background simple and the body unobstructed — the environment is added separately in step 4, and the video stage will move the character.
4. **Environment image** — the world the character lives in. Generate it as a separate image so the same location can hold several characters and props.
- `generate_image(model='t2i', prompt=..., orientation='landscape', resolution='1080P', name='<Environment name>', entity_type='locations')`
- **Real environment**: photographic realism — "real kitchen counter, real city street at dusk, real vegetable market stall, real park lawn, real subway platform, natural lighting, 35mm photo look, shallow depth of field, no people, no characters".
- **Artificial / stylized environment**: "studio stage with soft pastel backdrop and floor sweep, cartoon fantasy meadow, cozy animated wooden cabin interior, colorful cartoon city, seamless backdrop, 3D render look".
- Always ask for **no people / no characters** in the environment image — the characters are added by the video step.
5. **Prop images** (only when the scene has objects the characters handle) — one clean image per prop, photographed or rendered on a plain background so the video model can graft it into the scene.
- `gene