Updated: Apr 2, 2026
base model
RDBT [Anima]
Finetuned circlestone-labs/Anima.
Experimental, and will be updated randomly as needed.
This model is primarily designed to meet the high aesthetic level that humans can "draw". It does not provide overfitted default style. I use it as a starting point to stack more styles.
Dataset includes a wide range of images, common enhancements such as eyes, faces, hands, clothes, lightings, backgrounds, etc. All captions are natural language from Gemini.
Trained as a LoRA, for better training and distribution efficiency.
CFG distilled for better quality and stability.
Restrictions: Sharing merges using this model is not allowed. If you think this LoRA is useful, please share the link or the LoRA file.
If you are using a "custom" base model and this LoRA "breaks" the base model, instead of blaming the LoRA, try to look at the problem from a different angle: "Has this custom base model already merged this LoRA? And you merged it twice and the weights collapsed." Beware of fake base model creators. Some "creators" never do the training, they only grab other people's models, merge them, wipe all metadata and credits, and sell it as their own base model.
Usage:
Base model
This is a LoRA (a small patch of weights). You will need the base model (full weights) that trained the LoRA.
If you don't know what above means, just want to use a "finetuned base model", you can download this base model, which has merged this LoRA: https://civitai.com/models/2356447.
Prompt
Prefer natural language prompt. Prompt structure: style, subject, action, background.
You can omit all the quality tags. The average output quality of this model should be higher than so-called "masterpiece".
There are two trigger words:
"digital art", 2d art but not anime.
"digital anime illustration": 2d anime.
Recommended settings
"er_sde", "euler" or "euler_a" sampler.
20~30 steps.
cfg scale 1~3. Note: cfg 1 = disable negative prompt = 2x faster. Cover images are using cfg 1. cfg > 1 gives you better prompt adherence.
"Adjust Contrast" node to boost contrast.
FAQ
Why CFG distilled?
Stability and quality.
Distilled model has smooth sampling process, often has more natural and better details. e.g.: https://civitai.com/images/125705806 (Notice that the latent overflowed in many places during cfg sampling, no overflows in distilled model.)
But I want to use negative prompt to improve quality.
If you mean the general negative template "worst quality, score_1, jpeg artifacts, ...", then you don't have to. They have already been distilled out.
Update log
f = finetuned, d = cfg distilled.
Next (planned): 1536px training.
Next (TBD):
Bundled with AI styles + trigger words. (Open for suggestions)
DMD2. (What's the point, a 4-steps 2B model, looks cool, then what?)
Latest:
(3/28/2026) p2 v0.23fd: Rebased on preview2. Distillation: improved small details and stability (removed a regularization in distillation target and changed to second-order method).
Voting result: v0.20fd won. Thanks for the feedbacks.
(3/23/2026) p1 v0.20fd: Dataset: More furry. Finetuned base model: from v0.12 + more 100% steps. Better hands (?). Distillation: Fixed noisy pixels this time, really.
===============
Old:
(3/24/2026) preview1 v0.20fd b: Distillation: Different settings optimized for anime, high contrast and saturation.
(3/14/2026) preview v0.19fd b:
Updated dataset. Some private datasets have been dropped. You might notice the style changed.
Fixed high-freq artifacts in v0.12, now you should get a clear image without noisy pixels
b: Testing new distillation settings. Higher contrast. Aligned with common anime art.
(2/19/2026) preview v0.12fd:
Better stability and details, extended dataset.
(3/8/2026) preview v0.11fd 512px:
Prove of concept version for v0.12. Same dataset and settings as v0.12, except it was trained with 512px res.
Released by request, as it might be very useful. Running model in 512px and cfg1 is extremely fast (x10 faster, e.g. 30s -> 3s). If you don't have a beefy GPU. You can use this version to test your ideas/prompts in few seconds.
(2/12/2026) preview v0.6d:
CFG distilled only. No finetuning. Cover images are using Animeyume v0.1.
(2/3/2026) preview v0.2fd:
Speedrun attempt, mainly for testing the training script. Limited training dataset. Only covered "1 person" images plus a little bit of "furry". But it works, and way better than what I expected.