Pixal3D: Pixel-Aligned High-Fidelity Image to 3D Generation

Pixal3D Pixel-Aligned High-Fidelity Image to 3D Generation Tsinghua University & TencentARC SIGGRAPH 2026

What is Pixal3D? Pixal3D transforms 2D images into detailed 3D assets by explicitly projecting pixel features into a 3D volume, eliminating the spatial guesswork common in canonical-space generators. - Generates geometry and PBR materials - Supports single or multi-view input - Produces GLB meshes with PBR textures - Open-source under MIT license - Accepted to SIGGRAPH 2026

Architecture The key innovation is explicit pixel-aligned feature projection: instead of mapping images to a canonical 3D space (which loses spatial accuracy), Pixal3D projects pixel features directly into a 3D volume, preserving the exact spatial relationship from the input view. - Tech stack: Python, PyTorch, Trellis.2, Gradio

Applications - 3D content creation for games and VR - E-commerce product visualization - Digital twin generation - Rapid prototyping for 3D printing - 3D asset creation for film and animation

Get Started Website: https://pixal-3d.com/ GitHub: TencentARC/Pixal3D Hugging Face: TencentARC/Pixal3D Twitter: @staluxy Email: emluxy@gmail.com