Trained on Anima Base v1 using a carefully pruned dataset of 98 images with a combination of tags and natural-language captions. Trigger word: <code>@samon \(vxxr8575\)</code>.Recommended settings:<ul><li>30-40 steps</li><li>4-5 cfg</li><li><code>euler + beta57</code> or <code>er_sde + simple</code> . The <code>beta57</code> scheduler is better at bringing out the style's soft, painterly texture, but can cause melted hands and other anatomical issues.</li><li>❌ <code>euler_ancestral + beta57</code> will fry your gens. In fact, I ran into all kinds of artifacts with Euler A and other schedulers as well.</li></ul>Recommended resources:<ul><li><a rel="ugc" href="https://github.com/ClownsharkBatwing/RES4LYF">RES4LYF</a> for the <code>beta57</code> scheduler. Or just search "RES4LYF" in ComfyUI Manager.</li><li><a target="_blank" rel="ugc" href="https://github.com/sorryhyun/ComfyUI-Spectrum-KSampler">ComfyUI-Spectrum-KSampler</a>: training-free Anima acceleration. Shaves a solid 10 seconds off my inference time.</li><li>The <a target="_blank" rel="ugc" href="https://civarchive.com/models/950325/danboorue621-autocomplete-tag-lists-incl-aliases-krita-ai-support?modelVersionId=2444394">01.09.2025 version</a> of Danbooru/e621 autocomplete tag lists. Anima's knowledge cut-off date is September 2025, so this is the best version for Anima even though it's labeled as "ChenkinNoob 01".</li></ul>

samon_(vxxr8575) Style T-LoRA for Anima

<ul><li>Further pruned the dataset</li><li>Revamped the captions</li></ul>

Files with SHA256: f46734eeb30693c75868eb2105cf6149684317fe25cc7f60531ab4789cfac7e2