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
Trained more than 3x as long as v2, ending with 2 epochs at 896px to improve sharpness.
Resize is more conservative (sv_ratio=15 instead of 5), leading to larger filesize but outputs will be more faithful to the unresized model.
FAQ
Comments (6)
Will this work with the non-vpred version?
i dont understand what this lora dose. the programmer language is beyond me and my smooth brain lol
Besides the initial 1.5 base model and all photoreal models from that, the other types of models are either anime or furry, anime models are built from the danbooru dataset (danbooru is an image site that happens to use tags like best quality, masterpiece, low quality, etc) while furry models are built from e621 dataset which is a furry image site but that site doesn't really use quality tags so people need to prompt for specific artist names to get images that don't like average~bad quality.
This lora just makes "higher/better" image quality images using tags like "best quality"
Hello, can you by chance upload v2.1.1. I forgot to backup before a gitpull and it auto updated to v3.