NOTE: V4 changed the quality tags. Update your prompts.
This LoRA adds quality tags inspired by NAI to create aesthetically pleasing images without artist tags. The dataset was split into 4 tag groups based on their favcount on e621.
V4 is for fluffyrock-1088-minsnr-zsnr-vpred-ema (trained on e267), which requires a corresponding yaml file. You will also need the CFG Rescale extension.
V3 is for fluffyrock vpred (trained on e160).
Usage: add the tags to the prompt, use emphasis to change the strength of the effect. Do NOT use LoRA weight.
Positive prompt: (best quality, good quality:1.2)
Negative prompt: (worst quality, bad quality:1.2)Description
Initial release, resumed from a test lora with a smaller dataset, then trained with an improved larger dataset for 2 epochs at 512px then 1 epoch at 768px.
FAQ
Comments (3)
I've never used the model you trained this for, but the description has me curious-- what do the generations look like if you push this LORA too far into either direction?
Too much emphasis and it gets less and less saturated and coherent until it's just noise. Swapping positive and negative turns everything into bad ms paint doodles
@Feffy Like when one accidentally puts their negative prompt embeddings in the positive prompt? 😉
totally-have-never-done-thatCOUGH got-brain-bleach-on-speed-dialCOUGH 🥴