V0.2 Important Notes:
PSA: Expect humans to look uncanny.
Also, Anima was under-trained on Beastars characters, a character lora is recomended if you want them
Trained with Danboru tags and Natural language, real pain in the ass to manually write/edit but, better prompt adherence.
4000 Steps with 197 Images with Tags and 197 with NL
MFW my lora trainer shits the bed and corrupts 11 times after I painfully captured/tagged/edited 200 images
A style LoRA for Anima (by extension its finetunes) by CircleStone Labs.
This LoRA was trained on a curated dataset of 197 screenshot from the BEASTARS Season 1-2-3. All images were treated to remove text, and most symbols, then batch tagged with ConvNextV2 Huge Full and manually tagged for fixing/adding tags.
This Lora is best used at 1 strength, imo. Trigger word is very optional, but hey, its BE4STARS.
Also this was trained on my chef's kiss AMD 7900XT, (under Linux with Ostris's Ai-Toolkit❤️)
Description
Main Settings: 4000Steps, AdamW8bit, 1Batches, [768,1024], LR 4e-5, WeightDecay 1e-2, Rank32, Repeats 2(TAG) and 1(CAPTION), No X Axis Flip, 197 IMG (TAGS), 197 IMG (NL)
[Trained with both Danbooru Tags and Natural Language. Tags weight 1.4, Caption Weight 0.6]