<h2 id="rdbt-anima">RDBT [Anima]</h2>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: <a target="_blank" rel="ugc" href="https://civarchive.com/images/128870450">Original img</a>. <a target="_blank" rel="ugc" href="https://civarchive.com/images/128281978">Original prompt</a>.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/5ba29c17-afa2-46ce-b463-8885fd96485e/original=true/5ba29c17-afa2-46ce-b463-8885fd96485e.jpeg" />See Update Log section for version info. See <a target="_blank" rel="ugc" href="https://civarchive.com/models/2364703">this page</a> for original LoRA, <a target="_blank" rel="ugc" href="https://civarchive.com/models/2569103">this page</a> for PoC LoRA.All cover images are "raw" output, 1024px, no editing/upscale etc. Metadata included.Sharing merges using this model is not allowed.<hr /><h3 id="usage:">Usage:</h3>Settings (different from anima base model):<ul><li>Steps: 8, 12 or 24. Shift 3 or 1.</li><li>CFG scale: 1~4. Cover images are without CFG (CFG 1). You can enable CFG (CFG &gt;1) if you need higher prompt adherence (e.g. need higher style effect).</li></ul>PromptSpecific 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 "rough" trigger words, they are trained so they have effect, but they are not "specific style":<ul><li>@anime sketch: Low complexity. Rough outlines.</li><li>@digital anime illustration: Typical "anime". Clear and fine outlines. General complexity.</li><li>@digital art: More complex lighting, textures than typical "anime".</li><li>@cinematic digital art: More lighting, postprocess effects, semi-realistic, etc.</li></ul>FAQ: Base model built-in styles have less effect. This is a common problem for fringe content + distilled model + no cfg. You can:<ul><li>Enable CFG again.</li><li>Find/train a style LoRA.</li></ul>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.<hr /><h3 id="update-logs">Update Logs</h3>(5/12/2026): v0.32.b: Less step distillation (means higher diversity but less stability). 12 steps is still doable, 24 steps is recommended.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):<ul><li>"digital anime illustration": common 2d anime.</li><li>"digital art", 2d art but not anime, mostly digital art.</li><li>"anime sketch": simplified/unfinished anime drawing.</li></ul>(4/27/2026): v0.29: Distillation algorithm was almost completely rewritten.Increased diversity. Different seeds now can generate more different images. 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 <a target="_blank" rel="ugc" href="https://civarchive.com/models/2364703">this page</a>

RDBT | Anima

rdbtAnima_v032.safetensors

v0.32

v0.32.b

v0.29

v0.27

v0.25

p3 v0.24f 4-step dmd2

p3 v0.24f dmd2 turbo

p3 v0.24f 16-step dmd2

p1 v0.20f cfg

p1 v0.19f cfg b

p1 v0.12f cfg

fp16 patch v1.2

fp16 patch

p1 fp8

Files with SHA256: 64d41cbdd928f55c83d5203ddc304eff2f309a4ef4c85445290fbd767d1f363e