SD-WEBUI Extension for NUlIE 🐌

An algorithm based on decision trees and deep learning models to enhance images.
The algorithm is simple, it does not always provide good results on already great images.
The default maximum runtime is 30 seconds / image. The algorithm can provide good results in 10 seconds on some images.
The algorithm tries multiple transformations on contrast / saturation / lighting etc., it evaluates the changes and keeps the best ones.

Changelog:

• 2023/03/29: Added support for manual image editing (brightness, saturation, contrast..)
• 2023/03/28: Fixed issue on loading nn weights

Results / Original repository

• Standalone: NUlIE (with other results)

out2.mp4

Usage

After 2023/03/25

./webui.sh

• Navigate to the NUl Image Enhancer Tab

Before 2023/03/25

See: https://github.com/AUTOMATIC1111/stable-diffusion-webui/commit/af2db25c84c9a226ab34959e868fc18740418b4b
Before 2023/03/25, the SD-webui needed to be started with the following command:

./webui.sh --gradio-queue

/!\ Warning: This command was experimental and breaks the UI restart button.

Requirements

• sd-webui
• An NVIDIA GPU with > 2GB VRAM (very slow on CPU)
• CLIP
• safetensors

Models

• Aesthetic predictor: github.com/christophschuhmann/improved-aesthetic-predictor
• CLIP: github.com/openai/CLIP

➡️ More details on the algorithm: https://medium.com/p/how-i-built-an-ai-image-enhancer-f7d77186678e

⭐ If you like it! And I'll also try to improve this extension!