ComfyUI Facerestore TensorRT

This project provides an experimental Tensorrt implementation for ultra fast face restoration inside ComfyUI.

Note: This project doesn't do pre/post processing. It only works on cropped faces for now.

If you like the project, please give sa star! ⭐

⏱️ Performance

Note: The following results were benchmarked ComfyUI, using 100 similar frames

| Device | MODEL | PRECISION| FPS | |---------|--------|---|---| | RTX 3090 | Codeformer | FP16| 15.6| | RTX 3090 | Gfqgan | FP32| 13.1|

🚀 Installation

Navigate to the ComfyUI /custom_nodes directory

git clone https://github.com/yuvraj108c/ComfyUI-Facerestore-Tensorrt
cd ./ComfyUI-Facerestore-Tensorrt
pip install -r requirements.txt

🛠️ Building Tensorrt Engine

1. Download one of the following onnx models:

   • gfqgan.onnx
   • codeformer.onnx

2. Build tensorrt engines for these models by running:

   • python export_trt.py

3. Place the exported engines inside ComfyUI /models/tensorrt/facerestore directory

☀️ Usage

• Insert node by Right Click -> tensorrt -> Face Restore Tensorrt

🤖 Environment tested

• Ubuntu 22.04 LTS, Cuda 12.4, Tensorrt 10.4.0, Python 3.10, RTX 3090 GPU
• Windows (Not tested, but should work)

👏 Credits

• https://github.com/bychen7/Face-Restoration-TensorRT
• https://github.com/yuvraj108c/Codeformer-Tensorrt

License

Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0)