FUTURE EDIT!! THIS WILL BE UPDATED WITH GOOD (and manual) TAGGING AFTER I PLAN TO STOP MAKING >NEW< STYLE LORAS...

okay fixed the dataset (regular anime base model edition in version description)

fluffyrock does not require extensive negative prompting, and embeddings hurt it unless it's bad-hands-5

base prompt you should use for better quality images: sketch, artist name, signature, patreon logo, watermark, [bad-hands-5:0.7], realistic, flat colors, flat shading, simple shading, simple drawing, simple, reference sheet, (missing eyes, blank eyes)

if you think fluffyrock rikose (the one not in the version description) sucks, lower it down to 0.2-0.05 (lower TEnc to 0) cause it seems to just help with anatomy with artstyles (in fluffyrock) https://huggingface.co/lodestones/furryrock-model-safetensors/tree/main here are the fluffyrock models

i previously used "8_rikose and 8_eeveelutions" but lowered them in the dataset to 7 on both instead so it's less overtrained, if you are not a fan of the results, revert them back to 8

i don't recommend making so many folders like i did, do just one big one with all the artist's works or two if the artist drew something like pokemon a lot

these two things should help your results (for nai)

https://huggingface.co/datasets/gsdf/EasyNegative/tree/main

https://huggingface.co/FoodDesert/boring_e621/tree/main

tag how you want brother, newest version works at higher weights while the older ones work at lower

this is also my first lora and it took very long to set up, special thanks to argent and raven from the furry diffusion discord for helping get the trainer up and running though

i also highly recommend adding the anythingv4.vae/orangemix.vae to your results as that will just enhance the thing