tsunehimatoi/psd2live

What it solves

PSD2Live is an automated pipeline and desktop application designed to transform layered PSD files into fully rigged Live2D models. It eliminates the manual, labor-intensive process of creating meshes, setting up deformers, and configuring physics for 2D characters, allowing users to export editable .cmo3 projects and runtime .moc3 files with a single click.

How it works

The system processes PSD layers using a set of semantic labels (e.g., "front hair", "eyelash") to automatically identify character parts. It then applies several geometric and mathematical algorithms:

• Adaptive Mesh Generation: Uses Gaussian smoothing and Delaunay triangulation to create optimized meshes based on the image's alpha channel.
• Deformer Construction: Automatically builds a hierarchy of warps, including a "nine-axis" facial grid for head rotation (yaw, pitch, and roll) and specific constraints for eyes and mouths.
• Physics and Animation: Simulates hair sway and "jelly eye" dynamics using damped spring oscillators and generates a seamless 6-second idle animation.
• AI Agent Integration: Provides an MCP (Model Context Protocol) server that allows AI agents (like ChatGPT or Gemini) to read the project state, import assets, and edit keyframes or parameters via a streamable HTTP interface.

Who it’s for

Digital artists and VTuber creators who want to accelerate the Live2D rigging process without needing deep expertise in the Cubism Modeler software.

Highlights

• End-to-End Automation: From PSD input to .cmo3 and .moc3 export.
• Nine-Axis Facial Grid: Advanced mathematical modeling for realistic head movement.
• C1 Continuity: Ensures smooth curves for eyelid closure and mouth movements.
• MCP-Enabled Workspace: Allows external AI agents to assist in rigging and asset management.
• Official SDK Compatibility: Supports optional integration with the official Cubism SDK for pixel-perfect rendering verification.