OP:
This LoRA is for fluffyrock vpred (trained on e160).
The idea was to train a LoRA using quality tags similar to NAI to create aesthetically pleasing images without needing to specify tons of artist and style tags. The dataset was split into 5 groups based on their score on e621.
Usage example:
The effect isn't very strong by default, so I recommend using emphasis on the tags (NOT the LoRA weight!). I typically use weights of around 1.4.
Positive prompt: (best quality, high quality:1.4) Negative prompt: (worst quality, low quality, normal quality:1.4)