---
title: TransNormal-2 Demo
sdk: gradio
sdk_version: 5.49.1
app_file: app.py
python_version: 3.10.13
startup_duration_timeout: 1h
models:
  - Longxiang-ai/TransNormal-2
  - black-forest-labs/FLUX.2-klein-base-9B
tags:
  - computer-vision
  - monocular-normal-estimation
  - surface-normal-estimation
  - gradio
  - zerogpu
license: cc-by-nc-4.0
---

# TransNormal-2 Demo

Interactive Gradio demo for TransNormal-2 surface normal estimation.

Upload an RGB image, choose **Opaque** or **Transparent** scene mode, and run
the one-step FLUX.2 [klein] predictor followed by the Geometric Sharpening
Module (GSM).

## Runtime Notes

- This Space is intended to run on Hugging Face ZeroGPU first.
- Use ZeroGPU `xlarge` if available. The FLUX.2 [klein] 9B base model is large.
- Set `HF_TOKEN` in Space secrets after accepting the FLUX.2 [klein] license.
- First launch can take several minutes because the base model and TransNormal-2
  weights need to be downloaded and cached.

## Links

- Project page: https://longxiang-ai.github.io/TransNormal-2
- Code and README: https://github.com/longxiang-ai/TransNormal-2
- Model weights: https://huggingface.co/Longxiang-ai/TransNormal-2