defgsus / magic-pen

more experiments with ml image creation

This is supposed to become a painting app utilizing all the trippy neural image generation techniques. but i`m quite lacking time and GPU memory.

But here's a hack to run the stable-diffusion space on huggingface.co in a Qt application and save all images to a ./results/ folder:

pip install -r requirements.txt

python bin/qgenerator.py

• sub directory: is the path below ./results/ and can be empty
• filename slug: is the beginning of the filename and will be appended with increasing numbers
• prompt, negative prompt and guidance do the usual thing

You can create multiple tabs (Ctrl+T) for different prompts and request processing of the prompts multiple times at once. All jobs run in parallel ;-)

It's possible to quickly delete (D) or rate images (0 to 5) and filter the list for specific ratings.

The prompt and the rating is stored in separate json files along with each image.