trellis_blender

# TRELLIS Blender Plugin **Update: support text-to-3d and add mcp integration** A Blender addon that integrates [TRELLIS](https://github.com/microsoft/TRELLIS)'s 3D generation capabilities into blender. [TRELLIS](https://github.com/microsoft/TRELLIS) is SOTA text-to-3d and image-to-3d AIGC model.  ## Core Features * Text-to-3D: text -> textured 3D mesh * Image-to-3D: image -> textured 3D mesh * Text-conditioned Detail Variation: text + 3D mesh -> textured 3D mesh * Image-conditioned Detail Variation: image + 3D mesh -> textured 3D mesh * MCP integration: Integrates with MCP and can communicate with Cursor/Windsurf. Refer to [Trellis MCP](https://github.com/FishWoWater/trellis_mcp) ## Installation ### Requirements - Blender 3.6.0 or higher - Running TRELLIS API server (Refer to [my TreLLIS fork](https://github.com/FishWoWater/TRELLIS/blob/dev/README_api.md)) ### Enable the plugin 1. Download the plugin files (clone this repo) 2. In Blender, go to Edit > Preferences > Add-ons 3. Click "Install" and select the `trellis_for_blender.py` file 4. Enable the addon by checking the box next to "3D View: TRELLIS" ### Native Workflow 1. Access the panel (open 3d viewport and press n to open sidebar), find TRELLIS. 2. (Optionally) Select an object in the scene, select an image as conditional input. 3. Adjust generation parameters (empirically can use fewer steps) 4. The plugin will upload both the object(if selected) and the image/text to the API backend. 5. When finished, the model can be downloaded and imported into the scene. You can see the historical requests in the main panel ### MCP Workflow 1. Access the panel (open 3d viewport and press n to open sidebar), find TRELLIS, click `Start MCP Server` 2. Open Claude/Cursor/Windsurf, paste following configuration: ``` text "mcpServers": { "trellis": { "command": "uvx", "args": [ "trellis-mcp" ] } } ``` ## Generation Parameters - **Sparse Structure Settings** - Sample Steps (sampling steps for structure diffusion, by default 12) - CFG Strength (classifier-free-guidance, by default 7.5, higher value will better align the input image) - **Structured Latent Settings** - Sample Steps (sampling steps for SLAT diffusion, by default 12) - CFG Strength (classifier-free-guidance, by default 7.5, higher value will better align the input image) - **Postprocessing Mesh Options** - Simplify Ratio (# of triangles to remove, by default 0.95) - Texture Size (by default 1024, can set to 2048 for higher quality, but slower) - Texture Bake Mode ('fast' or 'opt', 'opt' can be slow but has higher quality) ## Features - Asynchronous request processing - Real-time status updates - Error handing * "No selected file": Select an input image * "API connection error": Check if the API server is running * "Processing error": Check the API server logs for details Any issue/discussion/contribution is welcomed!