Anima 3.8B z3v0.1 is a style merge on LyloGummy's 52-block Anima 3.8B (the BASE / preview release, which pairs the DiT with a Qwen3.5 4B cross-attention adapter). Six Anima LoRAs — five style LoRAs plus an eye-detail LoRA — were remapped from 28 to 52 blocks and baked in at fixed weights, then the model was quantized to MXFP8. It is the 3.8B sibling of <a target="_blank" rel="ugc" href="https://civarchive.com/models/3007132">Anima 2B z3v0.1</a> and <a target="_blank" rel="ugc" href="https://civarchive.com/models/3007188">Anima 2.9B z3v0.1</a> and carries the same five styles, tuned for this architecture.Compared with the 2B and 2.9B versions it follows longer natural-language prompts and multi-subject instructions better, and renders cloth and hair more realistically; the line art is softer.<h3 id="requirements-(read-first)">Requirements (read first)</h3>This is not a plain Anima checkpoint. You need all of:<ul><li>ComfyUI 0.33.1 or newer (52-block support and MXFP8 quantized layers; a GPU with MXFP8/FP8 support is recommended — tested on an RTX 5080)</li><li>Custom node comfyui-anima-3-8B (the original v1 / preview node for Anima 3.8B BASE — not the v1.1 bundled-connector release)</li><li><code>Anima-3.8B-expanded_adapter.safetensors</code> (the Qwen3.5 adapter from the Anima 3.8B BASE release) in <code>models/text_encoders</code></li><li><code>qwen35_4b.safetensors</code> (Qwen3.5 4B text encoder) in <code>models/text_encoders</code></li><li><code>qwen_3_06b_base.safetensors</code> (native Anima text encoder) in <code>models/text_encoders</code></li><li><code>qwen_image_vae.safetensors</code> in <code>models/vae</code></li><li>About 12–13 GB of VRAM at 896x1152 in our tests</li></ul>Base model and adapter: <a target="_blank" rel="ugc" href="https://civarchive.com/models/2880480?modelVersionId=3255578">https://civarchive.com/models/2880480?modelVersionId=3255578</a> and <a target="_blank" rel="ugc" href="https://huggingface.co/lylogummy/Anima-3.8B">https://huggingface.co/lylogummy/Anima-3.8B</a> . Qwen3 0.6B and the VAE: <a target="_blank" rel="ugc" href="https://huggingface.co/circlestone-labs/Anima">https://huggingface.co/circlestone-labs/Anima</a> .<h3 id="recommended-settings">Recommended settings</h3><ul><li>Steps: 30</li><li>CFG: 4.5 (negative prompts work)</li><li>Sampler / scheduler: <code>res_multistep</code> / <code>beta</code></li><li>Model sampling: none (no ModelSamplingAuraFlow node)</li><li>Resolution: about 1 megapixel, e.g. 896x1152 (portrait), 1152x896 (landscape), 832x1216</li><li>Do not add a turbo/distill LoRA and do not load the baked LoRAs again. Ordinary 28-block Anima LoRAs must be remapped to 52 blocks before they work.</li></ul><h3 id="prompting">Prompting</h3><ul><li>No trigger words needed. Danbooru-style tags plus a short plain-English paragraph (scene, light source, mood) work best; the showcase prompts use exactly this pattern.</li><li>Tip: start the prompt with <code>enchanting eyes</code> — it activates the baked eye-detail LoRA (richer irises and lashes).</li><li>For several characters, give each one its own sentence and use names instead of pronouns.</li><li>Close and half-body framing gives the sharpest faces; in wide shots faces get soft.</li><li>Add <code>shiny skin, oily skin</code> to the negative if skin gets too glossy; <code>adult woman, mature female</code> plus <code>child, loli, chibi</code> in the negative for adult proportions.</li><li>For full-body outfits, state the skirt length; hair tends to come out voluminous.</li></ul><h3 id="comfyui">ComfyUI</h3><code>UNETLoader</code> (file in <code>models/diffusion_models</code>, weight dtype default) -&gt; sampler. Text: <code>CLIPLoader</code> with <code>qwen_3_06b_base.safetensors</code> (type <code>stable_diffusion</code>) + <code>Load Qwen3.5 4B (Anima)</code> with <code>qwen35_4b.safetensors</code> + the Anima 3.8B adapter, all into the Anima 3.8B unified prompt node; use its <code>expanded</code> output as conditioning. <code>VAELoader</code> with <code>qwen_image_vae.safetensors</code>. Do not set the UNET loader to fp8 (the adapter path will crash); the MXFP8 layers load per-layer automatically.<h3 id="baked-components-(credits)">Baked components (credits)</h3><ul><li>Anima 3.8B, BASE (base) by LyloGummy, weight 1.0: <a target="_blank" rel="ugc" href="https://civarchive.com/models/2880480?modelVersionId=3255578">https://civarchive.com/models/2880480?modelVersionId=3255578</a></li><li>ushikani_kassen-styles, anima-base1.0 by zhihu, weight 0.6: <a target="_blank" rel="ugc" href="https://civarchive.com/models/2760349?modelVersionId=3106457">https://civarchive.com/models/2760349?modelVersionId=3106457</a></li><li>448 - guanren79145223's Style, ANIMA by monkewt, weight 0.5: <a target="_blank" rel="ugc" href="https://civarchive.com/models/2765265?modelVersionId=3112478">https://civarchive.com/models/2765265?modelVersionId=3112478</a></li><li>CowCowCrab / Saiou-ga-ushi Style, v4.0 (Anima) by yeetyah145, weight 0.5: <a target="_blank" rel="ugc" href="https://civarchive.com/models/1757934?modelVersionId=3165356">https://civarchive.com/models/1757934?modelVersionId=3165356</a></li><li>@zxcvnmweruo Style Anima, v1.0 by Umbre99, weight 0.7: <a target="_blank" rel="ugc" href="https://civarchive.com/models/2830588?modelVersionId=3193979">https://civarchive.com/models/2830588?modelVersionId=3193979</a></li><li>[Anima]ppolar, v1.0 by ScarletDimension, weight 0.7: <a target="_blank" rel="ugc" href="https://civarchive.com/models/2705088?modelVersionId=3037966">https://civarchive.com/models/2705088?modelVersionId=3037966</a></li><li>Enchanting Eyes, Anima by EmilyMars, weight 0.8: <a target="_blank" rel="ugc" href="https://civarchive.com/models/974076?modelVersionId=3027111">https://civarchive.com/models/974076?modelVersionId=3027111</a></li></ul>All LoRAs were remapped 28-&gt;52 blocks (existing blocks only; the 24 inserted blocks were left untouched). Thanks to the authors above, to LyloGummy for Anima 3.8B, and to CircleStone Labs for Anima.<h3 id="license-and-use">License and use</h3>Non-commercial. Derivative of Anima (CircleStone Labs Non-Commercial License), of Anima 2.9B / Anima 3.8B, and of the components above; their terms apply. Do not sell the model or offer it as a paid generation service.<h3>中文简介</h3>Anima 3.8B z3v0.1 是基于 LyloGummy 的 52 层 Anima 3.8B（BASE / 预览版，配 Qwen3.5 4B 适配器）的画风融合模型：把五个画风 LoRA 和 Enchanting Eyes 眼部 LoRA 从 28 层重映射到 52 层后按固定权重烤入，再量化为 MXFP8。与 Anima 2B / 2.9B z3v0.1 同一套画风，更听长句和多人描述，布料和头发更写实，线条偏软。需要 ComfyUI 0.33.1+、comfyui-anima-3-8B（v1 预览版节点）、Anima 3.8B 适配器、qwen35_4b、qwen_3_06b_base 和 qwen_image_vae。推荐 30 步、CFG 4.5、res_multistep + beta、约 100 万像素，不加 shift。不需要触发词；提示词开头加 <code>enchanting eyes</code> 可以让眼睛更细致。仅限非商用。

Files with SHA256: 3be6121b30109b1ac7fc6b599e35589219459f9a7b038f4a2ef6a03182761318