<h1 id="anima29b-turbo-v2.9-lora-(unofficial-block-remap)">Anima29B Turbo v2.9 LoRA (Unofficial Block Remap)</h1><p><strong>Unofficial community derivative LoRA</strong>: a <strong>block-index remapped</strong> version of Turbo-ANIMA-v2.9 for Anima-2.9B. NOT an official release by Anima / CircleStone Labs.</p><p><strong>Use with the original Anima-2.9B Preview 1</strong>: gives the 2.9B base turbo speed (8 steps) while keeping its quality, character/artist knowledge and NSFW performance.</p><ul><li><p>File: <code>anima29b-turbo-v2.9-remap.safetensors</code></p></li><li><p>Size: ~227 MiB — LoRA (written directly in the 40-block layout)</p></li></ul><h2 id="why-the-remap">Why the remap</h2><p>Anima 2.9B is a 28→40 block depth extension of Anima base (new blocks inserted at indices 2,5,8,11,14,17,21,24,27,30,33,36). The original Turbo-ANIMA LoRA is in the 28-block layout and <strong>silently misaligns</strong> when attached to 2.9B (a runtime patch is otherwise required). This remapped version has its keys rewritten offline to the 40-block layout — it loads correctly <strong>without any runtime patch</strong>. Verified pixel-identical against "original LoRA + runtime patch" under the same seeds (6 scenes incl. NSFW).</p><h2 id="workflow">Workflow</h2><p>UNETLoader (<code>anima29B_v10.safetensors</code>, the original Anima-2.9B Preview 1) → <strong>LoraLoaderModelOnly</strong> (this LoRA, strength <strong>0.8</strong>) + CLIPLoader (<code>qwen_3_06b_base.safetensors</code>) + VAELoader (<code>qwen_image_vae.safetensors</code>) → two CLIPTextEncode → KSampler → VAEDecode → SaveImage.</p><h2 id="recommended-parameters">Recommended parameters</h2><ul><li><p>LoRA strength <strong>0.8</strong> (1.0 usable; 0.5 under-converged, unusable)</p></li><li><p>steps <strong>8</strong> (8-12)</p></li><li><p>sampler <strong>euler + sgm_uniform</strong></p></li><li><p>cfg: <strong>1.0</strong> without negative; <strong>1.2</strong> with negative (range 1.0-1.6)</p></li><li><p>resolution: official 812×1216 / 1152×1536 / 1536×1536; 16-multiples like 848×1264 tested fine (~11-19 s/image)</p></li></ul><h2 id="prompting-(no-quality-tags)">Prompting (NO quality tags)</h2><p>The 2.9B base has no quality/score tags in its training data — <strong>do not use</strong> masterpiece / best quality / score_N:</p><pre><code>year 2026, &lt;safe/sensitive/nsfw/explicit&gt;, &lt;count&gt;, &lt;character (series)&gt;, @&lt;artist&gt;, &lt;appearance tags&gt;, &lt;natural language sentences (2-3+)&gt;</code></pre><ul><li><p>Keep the <code>year 2026</code> tag; character right after series; avoid underscores</p></li><li><p>An explicit <code>@artist</code> tag is recommended to anchor the style; write 2-3+ full natural-language sentences</p></li><li><p>Negative generally stays empty (cfg 1.0); against fake signatures/watermarks write <code>text, letters, sign, logo, watermark, signature, username, artist name</code> + cfg 1.2</p></li><li><p>For NSFW just write the rating and content tags; no "uncensored"-style positive tags needed</p></li></ul><h2 id="sample-images">Sample images</h2><p>Samples are straight outputs of this LoRA at strength 0.8, 8 steps, cfg 1.0. Full ComfyUI workflow metadata is embedded in the PNGs — drop into ComfyUI to reproduce.</p><h2 id="related">Related</h2><ul><li><p>Sister project (full model): <strong>Anima29B P1 + Turbo 1.1 Merged Model</strong> (<code>anima29b-p1-turbo1.1</code>) — no LoRA needed, quality tags work, turbo prompt style</p></li><li><p><a target="_blank" rel="ugc" href="https://github.com/Langzaigg/ComfyUI-AnimaFlow">ComfyUI-AnimaFlow</a>: AI tool-use API (MCP / HTTP) for Anima-family models, ships a ready-made workflow (<code>anima29_turbo</code>) and prompt knowledge for this LoRA</p></li><li><p>Upstreams: <a target="_blank" rel="ugc" href="https://civarchive.com/models/2619830/turbo-for-anima-less-steps">Turbo-ANIMA</a>, <a target="_blank" rel="ugc" href="https://huggingface.co/Gazingstars123/Anima-2.9B">Gazingstars123/Anima-2.9B</a>, <a target="_blank" rel="ugc" href="https://huggingface.co/circlestone-labs/Anima">circlestone-labs/Anima</a></p></li></ul><h2 id="license">License</h2><p>All upstreams (Turbo-ANIMA / Anima-2.9B / Anima) use the <strong>CircleStone Labs Non-Commercial License</strong>; as a derivative work this remapped LoRA follows the same license: <strong>non-commercial use only</strong>, and the license notice must be preserved on redistribution.</p><hr /><h1 id="anima-2.9b-turbo-v2.9-lora-unofficial-remap">Anima 2.9B Turbo v2.9 LoRA（非官方块重映射版 Unofficial Remap）</h1><p><strong>非官方（unofficial）社区衍生 LoRA</strong>：Turbo-ANIMA-v2.9 的 Anima-2.9B <strong>块索引重映射（remap）</strong>版本，非 Anima / CircleStone Labs 官方发布。</p><p><strong>配合原版 Anima-2.9B Preview 1 使用</strong>：给 2.9B 底座加上 turbo 速度（8 步），底座模型的画质、角色/画师知识与 NSFW 表现不变。</p><ul><li><p>文件：<code>anima29b-turbo-v2.9-remap.safetensors</code></p></li><li><p>大小：约 227 MiB —— LoRA（40 块布局直写）</p></li></ul><h2>为什么需要重映射版</h2><p>Anima 2.9B 是 base 的 28→40 块深度扩展（新块插入在索引 2,5,8,11,14,17,21,24,27,30,33,36）。原版 Turbo-ANIMA LoRA 是 28 块布局，直接挂到 2.9B 上键索引会<strong>静默错位</strong>（需要运行时补丁动态修正）。本重映射版已离线把键索引改写为 40 块布局，<strong>无需任何运行时补丁</strong>即可正确加载生效——已与"原版 LoRA + 运行时补丁"做同 seed 逐像素对照（6 场景含 NSFW），输出完全一致。</p><h2>工作流</h2><p>UNETLoader（<code>anima29B_v10.safetensors</code>，即原版 Anima-2.9B Preview 1）→ <strong>LoraLoaderModelOnly</strong>（本 LoRA，strength <strong>0.8</strong>）+ CLIPLoader（<code>qwen_3_06b_base.safetensors</code>）+ VAELoader（<code>qwen_image_vae.safetensors</code>）→ 双 CLIPTextEncode → KSampler → VAEDecode → SaveImage。</p><h2>推荐参数</h2><ul><li><p>LoRA 强度 <strong>0.8</strong>（1.0 可用；0.5 欠收敛不可用）</p></li><li><p>steps <strong>8</strong>（可用 8-12）</p></li><li><p>采样器 <strong>euler + sgm_uniform</strong></p></li><li><p>cfg：不写负面 <strong>1.0</strong>；写负面 <strong>1.2</strong>（范围 1.0-1.6）</p></li><li><p>分辨率：官方推荐 812×1216 / 1152×1536 / 1536×1536，848×1264 等 16 倍数实测良好（约 11-19 秒/张）</p></li></ul><h2>提示词（不要质量词）</h2><p>2.9B 底座训练数据无质量/分数标签，<strong>禁用</strong> masterpiece / best quality / score_N 类标签：</p><pre><code>year 2026, &lt;safe/sensitive/nsfw/explicit&gt;, &lt;人数&gt;, &lt;角色 (作品)&gt;, @&lt;画师&gt;, &lt;外貌标签&gt;, &lt;自然语言句子 2-3 句&gt;</code></pre><ul><li><p>保留 <code>year 2026</code> 年份标签；角色名紧跟作品标签，避免下划线</p></li><li><p>建议显式给 <code>@画师</code> 锚定画风；自然语言写足 2-3 句</p></li><li><p>负面一般留空（cfg 1.0）；排除假签名/水印时写文字类负面 <code>text, letters, sign, logo, watermark, signature, username, artist name</code> 并配 cfg 1.2</p></li><li><p>NSFW 直接写分级与内容标签，无需额外正面"无码"类提示词</p></li></ul><h2>示例图</h2><p>示例为 LoRA 版（强度 0.8、8 步、cfg 1.0）直出。PNG 内含完整 ComfyUI 工作流，拖入即可复现。</p><h2>关联推荐</h2><ul><li><p>姊妹项目（完整模型版）：<strong>Anima29B P1 + Turbo 1.1 合并模型</strong>（<code>anima29b-p1-turbo1.1</code>）——不想挂 LoRA 时直接用，质量词有效的 turbo 提示词体系</p></li><li><p><a target="_blank" rel="ugc" href="https://github.com/Langzaigg/ComfyUI-AnimaFlow">ComfyUI-AnimaFlow</a>：Anima 系模型的 AI 工具调用 API（MCP / HTTP），内含本 LoRA 对应的开箱工作流（<code>anima29_turbo</code>）与提示词知识</p></li><li><p>上游：<a target="_blank" rel="ugc" href="https://civarchive.com/models/2619830/turbo-for-anima-less-steps">Turbo-ANIMA</a>、<a target="_blank" rel="ugc" href="https://huggingface.co/Gazingstars123/Anima-2.9B">Gazingstars123/Anima-2.9B</a>、<a target="_blank" rel="ugc" href="https://huggingface.co/circlestone-labs/Anima">circlestone-labs/Anima</a></p></li></ul><h2 id="license">License</h2><p>上游 Turbo-ANIMA / Anima-2.9B / Anima 均采用 <strong>CircleStone Labs Non-Commercial License</strong>，本重映射 LoRA 作为衍生作品遵循同一许可：<strong>仅限非商业用途</strong>，再分发需保留许可声明。</p>

Files with SHA256: 221f322d50ebaedf58dd0a67dd8c7830742942b2745e61842508dbded099285f