Skip to content

Latest commit

 

History

History
66 lines (49 loc) · 2.89 KB

README.md

File metadata and controls

66 lines (49 loc) · 2.89 KB

TRELLIS Blender Plugin

Update: I have deployed a ready-to-use endpoint url, so you can download the plugin and use it directly, NO need to deploy it by yourself:) But I use 3060 GPU and the low-vram mode, it can be a bit slow.

A Blender addon that integrates TRELLIS's 3D generation capabilities into blender. TRELLIS is SOTA text-to-3d(unreleased) and image-to-3d AIGC model.

DemoVideo

Core Features

Image-to-3D

  1. Access the panel (open 3d viewport and press n to open sidebar), find TRELLIS.
  2. Select an image as input.
  3. Adjust generation parameters (empirically can use fewer steps)
  4. Click Image to 3D.
  5. The plugin uploads the image to API backend.
  6. When finished, the model url can be downloaded and imported into the scene.

DemoPic1

Image-Conditioned Detail Variation

can be used to synthesize texture for an existing 3D model in the scene

  1. Access the panel (open 3d viewport and press n to open sidebar), find TRELLIS.
  2. 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 and the image to the API backend.
  5. When finished, the model url can be downloaded and imported into the scene (with geometry unchanged and new texture)

DemoPic2

Installation

Requirements

  • Blender 3.6.0 or higher
  • Running TRELLIS API server (Refer to my TreLLIS fork)

Steps

  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"

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!