RDBT | Anima [LoRA]
This page contains LoRA files for advanced users who want to build the ckpt from scratch (the pretrained ckpt), and want to minmax everything including the strength of RDBT.
See main page for more info.
See this page for update log and version info.
Note: Stacking this LoRA on a random merged ckpt is not recommended. Usually those ckpts have already merged tons of LoRAs. You will overwhelming the ckpt and this LoRA will not work as expected.
If I didn't release the LoRA for new version, means
Update is small, and I don't know if it works.
I trained the new version as accumulated LoRA, Then I have a LoRA strength bar for new data/method. Then I can easily test it.
However releasing multiple accumulated LoRAs for one version is quite tedious and misleading for most users. So I just release a merged ckpt.
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.
Special trigger words and latent watermark included.
Description
FAQ
Comments (12)
Amazing! can you share more what trainer did you use and a curated dataset example?
I was using my own implements, but this should work
https://github.com/tdrussell/diffusion-pipe/pull/505
"curated dataset example" basically my art bookmarks
nice, any chance you'll ever publish your scripts?
I was going to release the training code, it was based on diffusion-pipe as well. But obviously that I was too late, and not necessary any more. lol
https://github.com/tdrussell/diffusion-pipe/pull/505
Preliminary thoughts about anima:
The base model, aka. cosmos p2, is a solid model.
anima starts to struggle with 3+ objects, probably because the text encoder was swapped to qwen3, and the model lost some high level reasoning ability. Which is worse than cosmos p2 and lumina.
Obviously undertrained.
Still as solid as f***.
anima https://civitai.com/images/119935075
netayumi lumina https://civitai.com/images/115198437
anima https://civitai.com/images/119935889
netayumi lumina https://civitai.com/images/115192487
Many of the prompts of the cover images were stolen from others. Thank you.
And please forgive me for forgetting to cite them.
awesome. definitely a first improvement on the base. looking forward to more stuff from you!
This is a insane jump in quality.
This looks to be very promising for how ell Anima takes to training.
The speed of 2080ti and 5080 is same, and much faster than that of 4080 and 3090.That's strange.
so what is it, lora or base model?
So... this 80MiB LoRA just casually turned Anima into one of the best (semi-)realistic image models when used without any style or quality tags? In addition to halving generation times with CFG 1.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.