Anima 2.9B compatibility: Before using the 2.9B versions (including Turbo), update your ComfyUI or Forge Neo installation to the latest version and restart it. Older versions may not load or run the model correctly.BF16 / INT8 versions: SilvermoonMix-Anima-Evolved v2.3_2.9B now has an additional INT8 + ConvRot version, quantized from this model's BF16 weights. Choose BF16 or INT8 from the download menu. BF16 retains the original precision, while INT8 reduces file size and VRAM use and may run faster on supported hardware. Image details may differ.ComfyUI LoRA compatibility: Legacy Anima 2B LoRAs may need block remapping when used with the 2.9B versions. Community options include:<ul><li><a target="_blank" rel="ugc" href="https://github.com/shin131002/ComfyUI-Anima-Remap">ComfyUI-Anima-Remap</a></li><li><a target="_blank" rel="ugc" href="https://github.com/sparklingcoffee777/ComfyUI-Anima-2.9B-loraPatch">ComfyUI-Anima-2.9B-loraPatch</a></li><li><a target="_blank" rel="ugc" href="https://github.com/INuBq8/ComfyUI-Anima-LoRA-ControlNet-Patch">ComfyUI-Anima-LoRA-ControlNet-Patch</a></li></ul>Please refer to each project's documentation and choose the solution that suits your workflow.Anima 2.9B 兼容性提示：使用 2.9B 版本（含 Turbo）前，请先将所用的 ComfyUI 或 Forge Neo 更新至最新版本，并在更新后重启。旧版本可能无法正确加载或运行模型。BF16 / INT8 双版本：SilvermoonMix-Anima-Evolved v2.3_2.9B 现新增 INT8 + ConvRot 量化版，由本模型的 BF16 权重转换而来，可在下载菜单中选择 BF16 或 INT8。BF16 保留原始精度；INT8 体积更小、显存占用更低，在支持的硬件上可能更快，出图细节可能有所差异。ComfyUI LoRA 兼容性：在 2.9B 版本上使用旧 Anima 2B LoRA 时，可能需要层号重映射。可参考以下社区方案：<ul><li><a target="_blank" rel="ugc" href="https://github.com/shin131002/ComfyUI-Anima-Remap">ComfyUI-Anima-Remap</a></li><li><a target="_blank" rel="ugc" href="https://github.com/sparklingcoffee777/ComfyUI-Anima-2.9B-loraPatch">ComfyUI-Anima-2.9B-loraPatch</a></li><li><a target="_blank" rel="ugc" href="https://github.com/INuBq8/ComfyUI-Anima-LoRA-ControlNet-Patch">ComfyUI-Anima-LoRA-ControlNet-Patch</a></li></ul>请自行查阅各项目说明，选择适合自己工作流的方案。<hr />Fine-tuned on Anima with my "Evolved" training strategy. This strategy is good at:<ul><li>Stable hands</li><li>Neutral art style to work with artist tags</li><li>Multiple characters</li></ul>For instruction of using Anima, please refer to: <a target="_blank" rel="ugc" href="https://civarchive.com/models/2458426/anima">https://civarchive.com/models/2458426/anima</a>本模型基于 Anima，采用我的“Evolved”训练策略进行微调。该策略擅长：<ul><li>稳定的手部表现</li><li>中性画风，便于配合画师标签使用</li><li>多角色场景</li></ul>Anima 的使用说明请参考：<a target="_blank" rel="ugc" href="https://civarchive.com/models/2458426/anima">https://civarchive.com/models/2458426/anima</a><hr />Recommended settings:<ul><li>Sampling method: Euler a / ER SDE (Turbo)</li><li>CFG: 4 (2B) / 3 (2.9B) / 1 (Turbo)</li><li>Step: 20-28 / 8-12 (Turbo)</li><li>Start with no negative prompt. Sample images were generated without negative prompt.</li><li>Prefix artist with @. E.g. "@big chungus". You must put @ in front of the artist. The effect will be very weak if you don't</li></ul>Sample images were created with Forge Neo (<a target="_blank" rel="ugc" href="https://github.com/Haoming02/sd-webui-forge-classic/tree/neo">https://github.com/Haoming02/sd-webui-forge-classic/tree/neo</a>), featuring native Anima support. Right-click and open any sample image in a new tab to download the original file with its generation metadata.推荐设置：<ul><li>采样方法：Euler a / ER SDE (Turbo)</li><li>CFG: 4 (2B) / 3 (2.9B) / 1 (Turbo)</li><li>步数：20-28 / 8-12 (Turbo)</li><li>先尝试不使用负面提示词。示例图片都没有使用负面提示词。</li><li>在画师（或艺术家）名字前加上 @ 前缀。例如：“@big chungus”。你必须在画师名字前面加上 @，如果不这么做，效果会非常微弱。</li></ul>示例图片由原生支持 Anima 的 Forge Neo (<a target="_blank" rel="ugc" href="https://github.com/Haoming02/sd-webui-forge-classic/tree/neo">https://github.com/Haoming02/sd-webui-forge-classic/tree/neo</a>) 生成。右键并在新标签页中打开任意示例图片，即可下载带有生成参数的原图。

SilvermoonMix-Anima-Evolved [2.9B/2B]

V2.1 -&gt; V2.2<ul><li>Fixed an error introduced during the model distillation stage of V2.1.</li><li>Reverted the dataset modifications from V2.1, as they were originally implemented to compensate for the distillation error mentioned above.</li></ul>Includes all V2.1 improvements<ul><li>Generates more natural lighting and improved overall logic.</li><li>Fixed an issue in the training tool that could lead to overexposed results.</li></ul>

Files with SHA256: 38d13927b6a3b158b9ef5140f4fad2c78bcc281001b03f4ced73b850ee1d7950