⚠️ Experimental ⚠️
Ohh ye, Mikuneki style while testing 2 new programs I made. Augmentation size predictions by mathematical equations, and probabilistic augmentation of images. This uses 38 original images augmented to 100, the minimum orig for this LoRa to work excellently is 33. Use any Illustrious models, euler a or dpm++(any, except 2m sde), and use 0.8-1.2 LoRa weight. If you see artefacts or noisiness, it's my fault, sorry... I forgot to save the final downscale to .png instead of .jpg, which means the quality got worse by 2x or more due to jpg problems with "Photocopier Effect".
