<h1 id="rinka-uzuki-style-lora">Rinka Uzuki Style LoRA</h1>A style LoRA trained on AI-generated works by Rinka Uzuki.Base Model: Anima Base v1.0<hr /><h2 id="v2">V2</h2>Trigger: <code>@Rinka Uzuki</code>Recommendation:<ul><li>Weight: <code>0.9</code></li><li>Try formatting the prompt as suggested by Anima's official documentation<ul><li>i.e. <code>[quality/meta/year/safety tags] [1girl/1boy/1other etc] [character] [series] [artist] [general tags]</code></li><li>Example:<pre><code>masterpiece, best quality, absurdres, safe, 1girl, Vivian Banshee, Zenless Zone Zero, @Rinka Uzuki, solo, from above, cowboy shot, hands on ground, looking at viewer, sitting, light smile, daytime, blue flowers, flower petals, depth of field, legs together.
She is sitting with her legs together among blue flowers, viewed from above. Her hands are resting on the ground as she looks toward the viewer with a light smile. Flower petals are scattered around her, with one petal resting gently on her pantyhose.</code></pre></li></ul></li></ul><hr /><h2 id="v1">V1</h2>Style Trigger: <code>rinka_uzuki_style</code>Recommended Weight: <code>0.9–1.0</code>Every training caption includes a framing tag selected from: <code>portrait, upper body, cowboy shot, feet out of frame, foot out of frame, head out of frame</code>A <code>close-up</code> tag is added where applicable.It's also recommended to specify the number of people, since this information is included in every training caption (e.g. <code>1girl</code>).The LoRA was trained using strictly verified Danbooru-style tags, while still working well with natural-language prompts.

Rinka Uzuki Style LoRA (Anima)

<ul><li>Fully aligned with the recommended Anima training guidelines</li><li>Captions combine booru-style tags with natural language descriptions</li><li>Data cleanup<ul><li>Rebalance in terms of framing, posture, etc.</li><li>Newer posts</li><li>Removed older upscaled images</li></ul></li><li>Uses a VLM + tagger captioning pipeline:<ul><li><code>animetimm/caformer_b36.dbv4-full</code><ul><li>Used as a supplementary tagger</li></ul></li><li><code>HauhauCS/Gemma4-31B-QAT-Uncensored-HauhauCS-Balanced-MTP</code><ul><li>Used as the primary VLM captioner prompted with injected tags</li></ul></li></ul></li></ul>

Files with SHA256: 5061e4943a4f516624851177b4a51e8202dd18a9eed6319db2a73e20012cd050