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.
This model doesn't provide default style nor quality filter. It's for better stability, prompt adherence and LoRA compatibility. Some cover images look ridiculous; they're just for demonstration purposes.
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 (15)
idk why styles tags not working here is there any solution. i tried many @styles
for example? probably don't have enough training samples. @style with more than 200 samples on danbooru should work.
You can try to increase your CFG to 4 without using any turbo Lora. I initially had a problem at first, but since I increased it, the style that I wanted was much more prevalent.
I had 20 sampling steps and 4 CFG, then put the @artist tag at the very front with no periods and it worked like a charm.
@MeisterGoon thank u
what is ymv0.5 ? peak after peak btw
The DiT models have hella potential.
In theory you could have a model that just simply does not mess up.
Nonetheless, I've been trying "crack the code" inbetween the model and the sampler, and between the conditioning and the sampler.
Not many new breakthroughs, but older research (not by me)... 'latent forcing' works amazing for it.
If you're Comfy user, you can get that over on my profile.
Making this comment on this one because this is my favorite Anima model so far. The version I have is wonky on poses and anatomy. But I'm going to try latest version now, and likely I'll post my feedback as a reply here.
Alright ymv0.5 v0.39 is definitely more solid.
More anatomical accuracy per image, and I can't complaint.
ymv0.5 v0.39 and p3 v0.32 are incredible models! 🤯 Thanks for sharing them. 🙏 Your models work wonderfully with my LoRAs. ✨ Keep up the great work! 🚀
I don't know what kind of dark magic you used. My character LoRa on your RDBT models can reproduce 95% what the training images look like, without any overfitted issues. It's more accurate than Anima b1. Your stabilizer LoRA can do this magic too. I'm wondering, is this because of the "comprehensive NL captions"? Can you share your method of captioning images?
BTW, v0.39 is rock-solid. I can't complaint. Your model is the unique star in an ocean flooded with 1girl and shiny AI styles.
@ikekph5 python script. Send image to LLM, ask for a detailed caption...
@reakaakasky how to do that? also does it support if the image is nsfw?
@monicalucci I use my own script, but there are more user friendly tools on github, you can search for it.
Close-sourced LLMs can't handle nsfw. I mainly use qwen3.6 now. With system prompt it can handle suggestive contents very well. I don't know whether it can handle hardcore nsfw, I don't care.
@reakaakasky I’ve switched over to that also but im using it for dan style tags with a little nl. It’s a bit slower, but LLMs are finally reliable enough to ‘see’ images without hallucinating, even if they don't catch absolutely everything.
By far the best Anima model I've tried, changes styles like a dream. I'd happily have to put a little more effort into picking a style and have it be more responsive like this, great work!