Repolode
Submit
Supra-3D

Supra-3D/SUPRA-Tex

Generates textured, relightable 3D assets (GLB with base color, metallic, roughness maps) from a single image, and can also texture existing meshes.

View on GitHubWebsite
Stars
699
Forks
35
Open issues
4
Last commit
9 days ago

What is it?

SUPRA-Tex turns a single image into a 3D asset with physically based rendering (PBR) materials. It predicts materials in the reference view, encodes them as surface-aligned 3D observations, and uses an Interleaved Flow Transformer to fill in regions the camera cannot see. The goal is to keep fine details from the input, such as lettering, patterns, and material boundaries, on the visible surface.

The output is a GLB file with base color, metallic, and roughness textures, so it can be rendered under new lighting. Besides image-to-3D, the same pipeline can texture an existing mesh when you supply its camera, a foreground image, and reference-view material maps. A dual-view mode adds a generated back view so the unseen side also follows an image.

The project requires Linux, an NVIDIA GPU, and CUDA 12.4. It includes a local Gradio web app and command-line scripts. Some pipeline models, such as DINOv3 and RMBG-2, need separate access approval on Hugging Face.

Key features

  • 01Generates GLB assets with base color, metallic, and roughness textures.
  • 02Keeps lettering, patterns, and material boundaries from the input image on visible surfaces.
  • 03Completes textures for unseen regions with an Interleaved Flow Transformer.
  • 04Textures existing meshes given a camera, foreground image, and material maps.
  • 05Includes a local Gradio web app and command-line scripts.
  • 06Offers optional texture refinement and a depth pre-filter for occlusion boundaries.

Use cases

  • Creating textured 3D props or assets from a product or reference photo.
  • Texturing an existing mesh that already has geometry and camera data.
  • Producing PBR assets that need to be re-lit in a 3D engine or renderer.
  • Experimenting with image-to-3D generation on a single NVIDIA GPU.

Quick start

  1. 1

    Clone the repository

    git clone https://github.com/Supra-3D/SUPRA-Tex.git
  2. 2

    Install the runtime into the activated Conda environment (Python 3.10, CUDA 12.4 with nvcc required)

    python scripts/install_runtime.py --conda
  3. 3

    Log in to Hugging Face and get approved for DINOv3 access, then download the models

    python scripts/prepare_models.py
  4. 4

    Run the bundled example to generate a textured GLB

    python scripts/run_example.py --output outputs/spiny_whelk_shell.glb

Things to know

  • Requires Linux with an NVIDIA GPU and a CUDA 12.4 toolkit.
  • The image-to-3D pipeline depends on DINOv3, which needs gated access approval on Hugging Face.
  • RGB inputs without an alpha channel also need RMBG-2 or BiRefNet.

Topics

#3d-generation#pbr#texture-generation#mesh#image-to-3d#diffusion#cuda#3d#3dgeneration

FAQ

What is SUPRA-Tex?
Generates textured, relightable 3D assets (GLB with base color, metallic, roughness maps) from a single image, and can also texture existing meshes. SUPRA-Tex turns a single image into a 3D asset with physically based rendering (PBR) materials. It predicts materials in the reference view, encodes them as surface-aligned 3D observations, and uses an Interleaved Flow Transformer to fill in regions the camera cannot see. The goal is to keep fine details from the input, such as lettering, patterns, and material boundaries, on the visible surface.
Is SUPRA-Tex free to use? What license does it have?
SUPRA-Tex is open source under the MIT license. Check the license terms before using it in commercial products.
Is SUPRA-Tex still maintained?
Actively maintained. The last commit to SUPRA-Tex was on Oct 29, 2026.
What can I use SUPRA-Tex for?
Creating textured 3D props or assets from a product or reference photo. Texturing an existing mesh that already has geometry and camera data. Producing PBR assets that need to be re-lit in a 3D engine or renderer. Experimenting with image-to-3D generation on a single NVIDIA GPU.
How do I install SUPRA-Tex?
Clone the repository: git clone https://github.com/Supra-3D/SUPRA-Tex.git Install the runtime into the activated Conda environment (Python 3.10, CUDA 12.4 with nvcc required): python scripts/install_runtime.py --conda Log in to Hugging Face and get approved for DINOv3 access, then download the models: python scripts/prepare_models.py Run the bundled example to generate a textured GLB: python scripts/run_example.py --output outputs/spiny_whelk_shell.glb
Where can I get SUPRA-Tex?
The source code and installation instructions are on GitHub: https://github.com/Supra-3D/SUPRA-Tex

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects