gitmyhub

Pixal3D-ComfyUI

Python ★ 160 updated 7d ago

Pixal3D image-to-3D nodes for ComfyUI - local TencentARC Pixal3D generation with textured GLB export + Windows support

Pixal3D-ComfyUI is a plugin (called "custom nodes" in the ComfyUI ecosystem) that brings TencentARC's Pixal3D model into ComfyUI, a popular open-source tool for building AI image workflows. Pixal3D is an image-to-3D generation system: you give it a photo of an object, and it produces a 3D model in GLB format — a standard file format for 3D graphics that can be opened in tools like Blender, game engines, or viewed in browsers. The 3D model includes textures, meaning it looks like the original image and not just a blank shape.

ComfyUI lets users visually connect AI models together in a node graph, similar to a flowchart. This plugin adds Pixal3D as a set of nodes you can drop into that graph. Features include FlashAttention 2 and 3 support (a technique that speeds up AI processing), manual camera control for adjusting the viewing angle during generation, optional automatic background removal, and the ability to unload the model from GPU memory when not in use.

The setup is complex because Pixal3D requires several specialized CUDA libraries (compiled code for Nvidia GPUs) that must exactly match your Python version, PyTorch version, and GPU. The README includes detailed Windows-specific guidance with pre-built wheel files (installable packages) for common combinations. You would use this if you already work with ComfyUI for AI image generation and want to add 3D model creation from photos to your workflow. The full README is longer than what was provided.