RDBT [Anima]
Reinforcement learning distilled model. It delivers faster speed and higher aesthetics with only 12 NFEs, 5x faster than pretrained model (60 NFEs, 30 steps + cfg).
Example: Original img. Original prompt.
See Update Log section for version info. See this page for original LoRA, this page for PoC LoRA.
All cover images are "raw" output, 1024px, no editing/upscale etc. Metadata included.
Sharing merges using this model is not allowed.
Usage:
Settings (different from anima base model):
Steps: 8, 12 or 24. Shift 3 or 1.
CFG scale: 1~4. Cover images are without CFG (CFG 1).
Prompt
Specific style is required! This model does not provide a default style. You should always prompt specific style. Or use a style LoRA. Otherwise, you will get random/mixed style. This is a feature, not a bug. I use this model as a starting point to stack more style LoRA.
(v0.32+) There are some "roughly classified" trigger words, they are trained so they have effect, but they are not "specific style":
@anime sketch: Low complexity. Rough outlines.
@digital anime illustration: Typical "anime". Clear and fine outlines. General complexity.
@digital art: More complex lighting, textures than typical "anime".
@cinematic digital art: More lighting, postprocess effects, semi-realistic, etc.
Quality tags:
You can omit all the quality tags. Quality tags have been reinforced during distillation. Thus they don't have noticeable effects.
Same as negative tags. If you use cfg, there is no need to dump "score_1, blurry, worst quality, jpeg artifacts, extra arms,... x100 words" in your negative prompt. Those things have been distilled out.
Update Logs
(5/12/2026): v0.32.b: Less step distillation (means higher diversity but less stability). 12 steps is still doable, 24 steps is recommended for complicated prompt.
Styles reinforcement learning. I did this in v0.29, but not in v0.32.
(5/10/2026): v0.32:
No more green-ish, color shifting.
Second-order sampling, should have improved quality.
Trigger words have been reclassified to avoid model learning a unified style. See updated "Usage" section.
Old trigger words for backup (v0.29 and before):
"digital anime illustration": common 2d anime.
"digital art", 2d art but not anime, mostly digital art.
"anime sketch": simplified/unfinished anime drawing.
(4/27/2026): v0.29: Distillation algorithm was almost completely rewritten.
Increased diversity. This also improved lighting range, styles and LoRA compatibility.
Better details. This version can squeeze every single pixel out of the VAE.
(4/23/2026) v0.27: Improved stability, details.
(4/18/2026) v0.25: It's based on anima p3.
Previous testing versions, see this page
Description
FAQ
Comments (3)
What's the benefit of this over regular Anima? Just aesthetic?
speed
Fine-tuned checkpoints such as this generally perform better than the base model depending on what do you want to make: Realism, cartoonish, Anime style, etc. It is possible to get good results with the base Anima Preview-3 but its generally harder without adding stuff like "masterpice,best quality," and "low quality, bad hands" on the negatives, so that's why people usually look for fine-tunes to get better without having your prompts become a giant wall of text.
I have been trying this checkpoint and apart from the huge generation speed boost from generating at 1-3 cfg and 8-16 steps, it has been able to work well with an anime style Anima Preview-2 LoRA (haven't trained on Preview 3 yet) that I'm working on and it gets very nice results keeping the anime aesthetic from the LoRA with the added bonus of fast generation.