CivArchive
    SilvermoonMix-Anima-Evolved [2.9B/2B] - v1.0
    NSFW
    Preview 131355343
    Preview 131355396
    Preview 131355578
    Preview 131355587
    Preview 131355589
    Preview 131355597
    Preview 131355617
    Preview 131355635
    Preview 131355658
    Preview 131355677
    Preview 131355757
    Preview 131355764
    Preview 131355775
    Preview 131355778
    Preview 131355783
    Preview 131355791
    Preview 131355785
    Preview 131355789
    Preview 131355795

    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:

    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：

    请自行查阅各项目说明，选择适合自己工作流的方案。

    Fine-tuned on Anima with my "Evolved" training strategy. This strategy is good at:

    • Stable hands

    • Neutral art style to work with artist tags

    • Multiple characters

    For instruction of using Anima, please refer to: https://civarchive.com/models/2458426/anima

    本模型基于 Anima，采用我的“Evolved”训练策略进行微调。该策略擅长：

    • 稳定的手部表现

    • 中性画风，便于配合画师标签使用

    • 多角色场景

    Anima 的使用说明请参考：https://civarchive.com/models/2458426/anima

    Recommended settings:

    • Sampling method: Euler a / ER SDE (Turbo)

    • CFG: 4 (2B) / 3 (2.9B) / 1 (Turbo)

    • Step: 20-28 / 8-12 (Turbo)

    • Start with no negative prompt. Sample images were generated without negative prompt.

    • 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

    Sample images were created with Forge Neo (https://github.com/Haoming02/sd-webui-forge-classic/tree/neo), 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：

    • 采样方法：Euler a / ER SDE (Turbo)

    • CFG: 4 (2B) / 3 (2.9B) / 1 (Turbo)

    • 步数：20-28 / 8-12 (Turbo)

    • 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。

    • 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。

    示例图片由原生支持 Anima 的 Forge Neo (https://github.com/Haoming02/sd-webui-forge-classic/tree/neo) 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。

    Description

    FAQ

    Comments (16)

    ponzomut146May 21, 2026· 2 reactions
    CivitAI

    Is a pretty good checkpoint thank you!

    VKilkoMay 21, 2026· 1 reaction
    CivitAI

    Is the checkpoint also pretty good with NL and NL+Tag? All of your Sample images are Tag only.

    silvermoong
    Author
    May 22, 2026· 1 reaction

    Hi! I don't have many NL tags to test with on my end, but I've seen others getting positive results with them.

    bramfatura_of_earthMay 21, 2026· 3 reactions
    CivitAI

    Definitely a step in the right direction. Hands/Feet/Digits seems to be more stable compared to base, which is one of its pain points imo. Well done!

    MiracleKeyMay 21, 2026· 4 reactions
    CivitAI

    oh yes, anima model by my fav creator

    DeathspikeMay 22, 2026· 4 reactions
    CivitAI

    My first impression on this v1.0 model is extremely solid. It produces a clean style that I like. It does not need artist tags, but still supports them. It works well with a mix of tags and natural language. And it even works well with the preview3-based turbo LoRA. This may be the first Anima model I could see myself using as a daily driver. Well done!

    KoharuDeathPenaltyMay 23, 2026· 1 reaction
    CivitAI

    First impression mine is the best anima finetune so far ^^,great job man

    Neon_signsMay 24, 2026· 3 reactions
    CivitAI

    Really love this,idk how or why but I feel the prompt adherence is a little better with this

    XegakupMay 24, 2026· 3 reactions
    CivitAI

    Really love this
    it really stable - it can do things other mixes and even orig model can't

    chenNdGMay 24, 2026· 7 reactions
    CivitAI

    Hi @silvermoong thanks for this models ! It's amazing, did you plan to update some of you're LoRa to support Anima ?

    silvermoong
    Author
    May 24, 2026· 1 reaction

    Hi @chenNdG , yes, I can start making LoRAs on Anima

    fiendguilt616May 27, 2026· 5 reactions
    CivitAI

    Better than the base for most stuff, but the baked-in neutral style makes it hard to do unconventional stuff (eg. base understands perfectly what I mean when I type "2D flat line retro 90s anime style but made pixel-art", this barely does even with 2x focus on that part)

    cococolacakeJun 3, 2026· 5 reactions
    CivitAI

    Its really good.<3

    VKilkoJun 12, 2026
    CivitAI

    I really like your Model, when do you publish a new version?
    Do you can improve the work of negative prompt and add new characters in the next version?

    VKilkoJun 12, 2026

    By “negative prompt improvement,” I mean that there are certain things you simply can't effectively prohibit. For example, banning specific symbols in pupils or something specific to a character in a multi-character image.

    It would also be nice to further improve exact prompt following for complex prompts, such as specific directions, left-right specifications, and details for certain people, though it’s not bad as it is.

    silvermoong
    Author
    Jun 17, 2026

    Hi @VKilko , I have a new version in progress, but I'm still evaluating whether it's a significant enough improvement to release. My core training philosophy is to minimize the impacts to the base model's prompt adherence. So you likely won't see drastic changes in prompting.

    Checkpoint
    Anima
    by silvermoong
    Download (Beta) View on CivitAI
    base model

    Details

    Downloads
    3,191
    Platform
    CivitAI
    Platform Status
    Available
    Created
    5/21/2026
    Updated
    11/7/2026
    Deleted
    -

    Files

    qwen_image_vae.safetensors

    Size:
    242.05 MB
    SHA256:
    a70580f0213e67967ee9c95f05bb400e8fb08307e017a924bf3441223e023d1f

    Mirrors

    HuggingFace (353 mirrors)
    CivitAI (301 mirrors)
    TensorFiles (1 mirrors)
    ModelScope (3 mirrors)
    ModelScope CN (7 mirrors)

    SilvermoonMix-Anima1-Evolved-V1.safetensors

    Size:
    3.89 GB
    SHA256:
    490b5cf1ec4ffc2ca610214f0842daf7a05467a1606d6a8e0b7bf38568368bdd

    Mirrors

    HuggingFace (2 mirrors)
    CivitAI (1 mirrors)

    qwen_3_06b_base.safetensors

    Size:
    1.11 GB
    SHA256:
    cd2a512003e2f9f3cd3c32a9c3573f820bb28c940f73c57b1ddaa983d9223eba

    Mirrors

    HuggingFace (145 mirrors)
    CivitAI (267 mirrors)
    ModelScope (1 mirrors)
    ModelScope CN (2 mirrors)

    Available On (1 platform)

    Same model published on other platforms. May have additional downloads or version variants.

    SeaArt
    SilvermoonMix-Anima-Evolved - v1.0