Half-body Live2D Avatar Skill

English | 简体中文

Give an AI agent (such as Claude Code) one character illustration or a text description, and it builds a half-body Live2D avatar for streaming in VTube Studio, step by step. No manual modeling, no Cubism Editor, no mesh editing.

Features

• Head turns left/right and up/down (±30); the body leans when the head tilts; body follow and breathing
• Blinking and winking, eye tracking, eyebrow movement, lip sync, smiling mouth corners
• Hair physics in sections: bangs, side locks (root and tip swing separately), back hair or bun; twin tails each swing from the hair tie; long hair follows the body below the shoulders. Bangs are optional, and the side hair can be tucked behind the ears
• Necklace pendant swings with the body; drop earrings swing from the earlobes
• Glasses follow the face
• Hotkeys: smiling eyes, O-shaped mouth, angry, troubled; wave, open hands and pointing gestures; blush

Not supported: large head turns, full body, arms following the camera continuously.

The output is a model file VTube Studio loads directly (.moc3), not a Cubism Editor project, so the Editor can't open it. For production-grade results (finer physics, more expressions), import the separated layers into Cubism Editor and rig them again. Suited to front-facing half-body characters with short hair, hair past the shoulders or twin tails.

How the illustration is adjusted

The agent redraws your illustration or description into a version that separates cleanly into layers, and you approve it before the next step. The face, hair color, eyes, clothing, palette and art style stay the same; these are adjusted:

• Framing: front-facing half body from the top of the head to the waist, plain white background; arms hang at the sides with a gap from the body, hands empty. The three gestures are edited from this pose.
• Bangs: optional. With bangs, a straight-cut fringe ending above the eyebrows, both eyebrows fully visible so they can move; without, the whole forehead shows.
• Side hair: either one smooth section on each side of the face, covering the ear and ending at the chin, swinging on its own when the head turns; or tucked behind the ears, which stay visible.
• Long hair: hangs behind the neck, shoulders and back, showing only beside the neck and outside the arms; not over the shoulders, chest or arms.
• Twin tails: tied above and behind the ears, hanging behind the shoulders.
• Neck: a stretch of neck shows between the chin and the collar, so the chin slides over it when the head turns.
• Color and light: hair and skin in different colors, hair with a thin outline; flat lighting, no cast shadows. Shadows would shift out of place once the parts move.
• Accessories: glasses stay, drawn with a thin frame and clear lenses without glare; drop earrings stay, with the hair behind the ears; chokers and hanging ribbons are removed; a necklace stays only as a thin chain on the clothing with at most one pendant; hats sit on top of the head without covering the forehead or eyebrows.
• Blush: soft blush on the cheeks.

For hairstyles or accessories outside this list (animal ears, braids), ask the agent to extend the skill: add drawing rules, adjust the layer-splitting and rigging code, then check the result in VTube Studio.

What you do

1. Approve the illustration the agent generates.
2. Agree to upload it to See-through for layer separation (Hugging Face, ModelScope as backup; not needed when See-through runs on your Mac).
3. Check the avatar in VTube Studio against the checklist, then reply "pass" or "fail: what's wrong".

Requirements

• An AI agent that runs skills, such as Claude Code
• Python 3.10+, ffmpeg
• A ChatGPT account: images are generated with the Codex CLI (bundled with the ChatGPT desktop app). Another image tool that can edit images can be used instead (not yet tested on a full character)
• See-through, at least one of:
  • Hugging Face, free (preferred, a few minutes per run): create a Read token and put it in the HF_TOKEN environment variable or ~/.cache/huggingface/token. Without logging in, the daily free quota may not cover one run; logged in, about 1–2 runs a day
  • ModelScope, free (backup; queued, a run can take half an hour; higher resolution): create an access token and put it in MODELSCOPE_API_TOKEN or ~/.cache/modelscope/token
  • Or run See-through on your own Mac (optional; nothing is uploaded, no quota): tested on Apple silicon with 24 GB of memory, about 21 minutes per illustration. Setup in references/出图.md (“本机跑 See-through”)
• VTube Studio
• Live2D Cubism Core: included with Cubism Editor, or download the free Cubism SDK for Native

Tested on macOS. Not tested on Windows.

Getting started

1. Put this repository in your agent's skill folder. For Claude Code: git clone https://github.com/bellisji233/live2d-halfbody-skill ~/.claude/skills/live2d-halfbody
2. Install dependencies: pip install -r requirements.txt
3. Tell the agent "Make a VTube Studio avatar from this illustration" with the image path, or describe the character you want.

One character takes 19–24 generated images and one See-through run; from illustration to VTube Studio is about an hour, see FAQ.

FAQ

How long does one character take?

Timings from the third character (glasses):

Machine time adds up to about 30–45 minutes; with your approval and check, a character can be in VTube Studio within an hour if nothing goes wrong. The number of variants depends on whether the character has a hat or hair accessory, bangs, side hair in front of the ears, and pendants.

When the Codex or See-through quota runs out, the run stops at that step. Run it again once the quota resets; images already generated are not redone.

Language

The skill instructions (SKILL.md, references/) and script messages are in Chinese. The agent reads them as they are; talk to it in any language.

License

MIT, see LICENSE. rigkit/vendor/py-moc3 is MIT, license included. Live2D Cubism Core is not distributed with this skill; it is used under Live2D's license.