CivArchive
    AnimaPulse | Anima - v0.9
    NSFW
    Preview 131981215
    Preview 131981623
    Preview 131939903
    Preview 131974130

    [All AnimaPulse v1.x samples are generated using Diffucore UI. Try it today.]

    AnimaPulse a fine-tune of Anima.

    Fine-tuned with:

    • ~7.5k dataset for v0.1 and v0.2

    • 9.7k images or 19.4k NLP and booru tag dataset for v0.3

    • ~2.5k images or ~5k NLP and booru tag dataset for v0.9

    • ~7k dataset for v1.0

    • ~12.5k images or ~25k NLP and booru tag dataset for v1.1

    Newly Trained Knowledge Information:

    • v1.1: Arknights: Endfield from October 1st 2025 - June 20th 2026

    Recommended settings [Updated for v1.x]:

    • ER SDE or Secant Anneal [Diffucore exclusive]

    • simple or beta_mix [Diffucore exclusive]

    • 30-32 Steps

    • CFG 4.0-4.5

    This model is a fine-tuned (modified) version of the Anima model by CircleStone Labs. The Anima Model is licensed by CircleStone Labs LLC under the CircleStone Non-Commercial License. Copyright CircleStone Labs LLC. IN NO EVENT SHALL CIRCLESTONE LABS LLC BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH USE OF THIS MODEL.

    You may use Outputs (generated images) for commercial purposes, as stated on the CircleStone Non-Commercial License.

    As stated on the license discussion, I am allowed to ask for donations to cover model training expenses. Please donate to my Ko-Fi if you want to support future model training.

    Description

    Initial release that uses Anima v1.0 as base. Using 2.5k dataset with 2 stages, full NLP training and full booru tags training.

    FAQ

    Comments (9)

    dsadas2313May 27, 2026· 12 reactions
    CivitAI

    Nice work. Rather than simply taking other people's models, merging them without attribution, and passing them off as their own, creators who collect their own data and train their own models truly deserve credit.

    kayfahaarukku
    Author
    May 27, 2026· 1 reaction
    CivitAI

    Checking from my alt, sample images for v0.9 doesn't show up. Sorry for the inconvenience, should be showing up in an hour.

    wryman01May 27, 2026· 4 reactions
    CivitAI

    First impressions, this model is better at nsfw than most checkpoints, I don't have to use super high weights to get what I want

    Voodoo1029May 27, 2026
    CivitAI

    Is this an actual FFT or is this a lora merged in? Any chance you could share the settings you used for FFT? I'm considering doing one on a fairly large set and knowing some good values that have worked would be very helpful

    kayfahaarukku
    Author
    May 28, 2026· 1 reaction

    Hi, for this first 4 versions, everything are trained as a full matrix LoKr, then merged to the model. This method is rarely used but not new, firstly done with KohakuXL which later be the base of IllustriousXL.

    This method allows me to train with near FFT quality by only using a single 24GB consumer GPU.

    I plan to train the model as a FFT in the future, but have no funding for now.

    nogoMay 27, 2026
    CivitAI

    what kind of data is it trained on?
    just anime stuff or what?

    kayfahaarukku
    Author
    May 28, 2026

    Anime stuff, leaning more on NSFW and styles.

    notmyproblem324Jun 9, 2026
    CivitAI

    In the preview prompts there's a dot after score_7 is that intentional and if so is it any different than a comma or is it just for visibility?

    kayfahaarukku
    Author
    Jun 9, 2026· 2 reactions

    A typo 😅. Please don't overthink this too much. Generally should be a comma but it doesn't affect things very much.

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

    Details

    Downloads
    889
    Platform
    CivitAI
    Platform Status
    Available
    Created
    5/27/2026
    Updated
    11/11/2026
    Deleted
    -

    Files

    qwen_3_06b_base.safetensors

    Size:
    1.11 GB
    SHA256:
    cd2a512003e2f9f3cd3c32a9c3573f820bb28c940f73c57b1ddaa983d9223eba

    Mirrors

    HuggingFace (145 mirrors)
    CivitAI (275 mirrors)
    ModelScope (1 mirrors)
    ModelScope CN (2 mirrors)

    qwen_image_vae.safetensors

    Size:
    242.05 MB
    SHA256:
    a70580f0213e67967ee9c95f05bb400e8fb08307e017a924bf3441223e023d1f

    Mirrors

    HuggingFace (363 mirrors)
    CivitAI (310 mirrors)
    TensorFiles (1 mirrors)
    ModelScope (3 mirrors)
    ModelScope CN (7 mirrors)

    AnimaPulse-v0.9-gb-ep6.safetensors

    Size:
    3.89 GB
    SHA256:
    1c5d898701b9f68a3c7f476528a78e4d27e18a07a681d4bd688058a4fb7c4655

    Mirrors

    CivitAI (1 mirrors)