RDBT [Anima]
This is a general finetuned + distilled model.
Dataset contains ~10k handpicked images with accurate NL captions from LLM, body/hands anatomy. Does not contain any shiny plastic glossy AI image.
It improves the quality of the pretrained base model, from average danbooru level to the "masterpiece" level and beyond.
Some cover images look ridiculous; they're just for demonstration purposes for corner cases.
I use this model as a clean starting point to stack more LoRAs.
See this page for update log and version info.
For advanced users: RDBT model is trained as LoRA natively. See this page for original LoRA.
Base model:
prefix with ym: AnimaYume (hf link) (civitai link).
prefix with b (base), p (preview): Anima pretrained (hf link)
Sharing merges using this model is not allowed. This "restriction" won't affect anyone. It's only aimed at those who steal others' models to sell. If someone is selling this model as their own, I'm happy to list them here so everyone knows.
Known model thieves: NukeA.I (selling this model behind paywall on tensorart).
I wrote a story about it. Also contains a guide for trainers about "how to bake special trigger word into your model".
Usage:
Settings:
CFG: 1~3. This model has been distilled. You can disable CFG (CFG 1) and run the model 2x faster. Cover images are without CFG for demonstration. "RenormCFG" node is highly recommended if CFG is enabled (CFG > 1), set "renorm_cfg" value to 1.1.
Steps: 16+
Sampler: Euler (best diversity), Euler a/er_sde etc. (better stability)
Res: 1MP
Prompt:
Always specify style in prompt, or use a style LoRA. Otherwise, you will get random/mixed style. This is a feature, not a bug. This model does NOT have overfitted default style (which ignores prompt and is always active).
Quality tags:
Omit ALL quality tags. You don't need those. The fine-tuning dataset has higher quality than "masterpiece". Thus quality tags don't have effects. Omitting those redundant tokens allows LLM to pay more attention on other words.
Description
FAQ
Comments (9)
i also have left anima now, since krea 2 looks like a beast checkpoint
krea2 is still hyper filtered, anima is still better all around
"krea2 filter" is completely different than any other model filter.
Most models filter data before training.
Krea2 was trained on all kinds of things, and applied a safety filter (probably dpo?) afterwards. What the model learned is learned. Technically, this filter can be nuked. That's why people already so hyped.
@reakaakasky will your finetune restore like... *all* kinds of purely fictional content for krea2?
@RLG I'm not going to finetune krea2. My dataset is for quality, and krea2 don't need that.
v1 lets gooo
Thank for your work you really made the start of Anima a lot easier for me and motivated me for many lora with your checkpoint ! have fun with krea 2i'll look into it x) if you choose it
B1 V1 seems to be something of a downgrade as I am seeing extra fingers more often with less prompt adherence overall. Sometimes it even keeps/add things I either didn't prompt for or was actively trying to prompt out.
Edit: Also down to 3 fingers and a thumb in some images and lighting issues. IE: Daylight window in a scene prompted for night time
Idk what type of magic is this, but i just put CFG to 1, put bunch of shit i usually did with illustrious, and within 7 seconds ( with adetailer enabled ) it generates great pictures. Lol. Kind of sick of tinkering with anima checkpoints and waiting 30 secs just to generate 1 pic, so thanks a lot for your work bro.