DiffBIR offers two convenient ways to experience it:
Online demo (recommended for newbies)
Visit the official demo platform:https://replicate.com/zsxkib/diffbirYou can upload test images directly to experience super-resolution, denoising, and other features.
Complete process for local deployment
- environmental preparation::
git clone https://github.com/XPixelGroup/DiffBIR.git
cd DiffBIR
pip install -r requirements.txt - Model Download::
Get the pre-trained model (e.g. real_base or face_enhance) from the project page into themodels/catalogs - Running fixes::
python inference.py --input_dir inputs --output_dir outputs
Tip: The input image is recommended to be JPEG/PNG format below 512px, the output image is saved in outputs directory by default, and the whole process will take about 2-5 minutes (depending on the hardware configuration).
This answer comes from the articleDiffBIR: Intelligent Repair Tool to Improve Image QualityThe































