Mesh Avatar Studio Lets Claude Rig 2D VTuber Avatars Automatically

Mesh Avatar Studio turns a static illustration into an animated, VTuber-style 2D avatar. A coding agent handles mesh placement, layer cutting, eye and mouth variants, and initial rigging; a local editor provides the controls for refining the result.

The MIT-licensed project has attracted roughly 1,800 GitHub stars. It runs locally, renders through WebGL, and uses MediaPipe for face tracking. Finished avatars can blink, speak, turn, breathe, and move their hair. An experimental Rust runtime supports native playback.

The Agent Builds the Rig

Rigging converts a flat drawing into independently controlled regions such as the face, eyes, mouth, hair, and accessories. Mesh handles define how those regions deform as tracking values change. Preparing them manually requires careful coordinate work, layer masks, expression art, and repeated pose checks.

Mesh Avatar Studio gives that work to Claude Code or Codex through a repository guide. The agent checks coordinate accuracy on zoomed grids, places control points, cuts the source image into layers, and reviews fixed poses for deformation errors. Generated projects stay in the git-ignored projects/ directory.

The documentation recommends Claude Opus 5.5 with Claude Code or GPT-6.1 Sol with Codex. From a checked-out repository with its dependencies installed, Codex can run the pipeline with one command: