TUNET

A direct, pixel-level mapping from src to dst images via encoder-decoder.
Supports, training, inference or export to Compositor tools such Foundry Nuke or Autodesk Flame native inference.

🎥 Showcases:

AOVs:

https://youtu.be/TwvN8axWJLY
Models from the video are available for Nuke and Flame can be downloaded link below for test locally:
https://f.io/HovatFeX

Models has been trained in combination with Nvidia Cosmos foundation models.
Using 8x B200 GPUs. Inference can be done in consumer GPUs.

Rain:
Flame:

✅ Install using Miniconda or Anaconda: ###[Install Video)]

git clone --branch linux --single-branch https://github.com/tpc2233/tunet.git
cd tunet

conda create -n tunet python=3.10
conda activate tunet
pip install torch torchvision torchaudio 
pip install onnx pyyaml lpips onnxruntime Pillow albumentations PySide6 

For Windows or MacOS use the dedicated Branch:

check branches

✅ How to use, Open Tunet UI: ###[[Training Video soon]]

python ui_app.py

You are good to go!

✅ Tunet UI:

Main:

Training:

Converter:

Adv:

Data control:

Terminal CMD:

✅ Command line and terminal trainings are still available as usual, check docs:

soon

Video:

[video 🤗]

Citation

Consider cite TUNET in your project.

@article{tpo2025tunet,
  title={TuNet},
  author={Thiago Porto},
  year={2025}
}

License

The source code is licensed under the Apache License, Version 2.0. Commercial use Permission