README.md

---
title: SeqTex Texture Generator
emoji: 🎨
colorFrom: red
colorTo: yellow
sdk: gradio
sdk_version: 5.29.0
app_file: app.py
pinned: false
license: mit
short_description: Image-conditioned UV texture for GLB meshes via SeqTex
hardware: zero-gpu
models:
  - VAST-AI/SeqTex-Transformer
tags:
  - 3d
  - texture-generation
  - mesh
  - diffusion
---

# SeqTex Texture Generator

Upload an untextured **GLB mesh** and a **reference image** of your character's front view.

The pipeline will:

1. **Orient** the mesh to Z-UP and normalize it to unit scale.
2. **Unwrap UVs** with xAtlas if the mesh has none.
3. **Render** geometry position and normal maps in multi-view and UV space.
4. **Run SeqTex** to generate a full UV texture map conditioned on your image.
5. **Export** a downloadable textured GLB shown in an interactive 3D viewport.

## Recommended Space setup

1. In **Settings → Storage Buckets**, mount a bucket at `/data`.
2. In **Settings → Variables and secrets**, add `SEQTEX_SPACE_TOKEN` or `HF_TOKEN`.
3. Keep `AUTO_PREWARM=0` while debugging startup.
4. Open the **Cache / Startup** tab and click **Prepare cache now** after the UI loads.
5. Then use the **Generate Texture** tab.

## Notes

- The ZeroGPU call duration is capped at 120 seconds in `app.py`.
- `ninja` and `ninja-build` are included because `nvdiffrast` JIT-compiles a CUDA extension on first use.
- The Storage Bucket cache paths are `/data/hf_home`, `/data/hf_home/hub`, `/data/torch`, and `/data/torch_extensions`.

## Credits

Model: [VAST-AI/SeqTex-Transformer](https://huggingface.co/VAST-AI/SeqTex-Transformer)

Space utilities adapted from [VAST-AI/SeqTex](https://huggingface.co/spaces/VAST-AI/SeqTex).