Live2D Automation

You can generate complete Live2D models automatically from a single photo using a dedicated MCP server. This server handles AI-based segmentation, layered Live2D generation, rigging, physics configuration, and motion creation, giving you an end-to-end pipeline that you can run locally or integrate into your tools.

How to use

You will run the MCP server locally and connect to it with an MCP client. Start by launching the server, then use its tools to analyze photos, generate layers, rig the model, configure physics, and create motions. You can also invoke a full pipeline programmatically to produce a complete Live2D model and its motions in one go.

Start the MCP server from your project directory using Python. This runs the server module that exposes the MCP endpoints for the client to interact with.

How to install

cd live2d_automation
pip install -r requirements.txt

Additional usage notes

You can also generate a complete Live2D model in a single step by calling the full pipeline from code. This demonstrates a typical end-to-end run that starts from an input photo and outputs the model and motions to a directory.

# A complete end-to-end pipeline call
# Ensure this is run in a Python environment where the package is installed
from live2d_automation.mcp_server.server import full_pipeline

result = await full_pipeline(
    image_path="path/to/photo.png",
    output_dir="output/",
    model_name="MyCharacter",
    motion_types=["idle", "tap", "move", "emotional"]
)

Output structure

The pipeline outputs a structured directory containing the model configuration, textures, and motion data. A typical layout looks like this:

output/
└── [model_name]/
    ├── model3.json       # Model configuration
    ├── physics.json      # Physics settings
    ├── textures/         # Textures for layers
    │   ├── layer_head.png
    │   ├── layer_body.png
    │   └── ...
    └── motions/            # Motion files
        ├── Idle_Breath.motion3.json
        ├── Idle_Blink.motion3.json
        ├── Tap_Head.motion3.json
        └── ...

System requirements

Ensure you have a supported Python environment and an appropriate compute capability for the model processing. The recommended setup includes Python 3.8 or newer and a capable GPU if you plan to accelerate processing.

License

MIT license applies to the MCP server.

How to run the MCP server (direct start)

Run the MCP server directly from your project to start listening for client requests.

python -m live2d_automation.mcp_server.server

Notes on using with an MCP client

Use your MCP client to connect to the server, then access the available tools to analyze photos, generate layers, create meshes, set up rigging, configure physics, and generate motions. The client can also invoke the full_pipeline to produce a complete model and motion set in one go.

Available tools

analyze_photo

Analyze a photo to detect character pose and outline data that informs subsequent steps.

generate_layers

Create layered Live2D textures and meshes from analyzed data.

create_mesh

Build ArtMesh grids for accurate deformation and rendering.

setup_rigging

Bind bones and rigging for motion and deformation control.

configure_physics

Configure physics simulations for realistic movement and interactions.

generate_motions

Produce motion files for idle, blinking, tapping, and other actions.

full_pipeline

Run the complete one-click pipeline from photo to model and motions.