CivArchive
    Preview 146045089
    Preview 146045152
    Preview 146045022
    Preview 146045015
    Preview 146045318
    Preview 146045072
    Preview 146045105
    Preview 146045049
    Preview 146045120
    Preview 146045175
    Preview 146045237
    Preview 146045256
    Preview 146045228
    Preview 146055986
    Preview 146056170

    anima UWU V2


    The model has been fine-tuned on over 36,000 images from Danbooru, spanning the years 2024–2026. I curated this dataset from a larger collection.


    Training was carried out using adapters.


    Yes, an adapter—LoRA is what you use when you have 200–500 annotated images. But when you have tens of thousands of images containing dozens or even hundreds of different concepts and styles, that’s a full-fledged adapter that complements the model.

    anima UWU Base


    This stylistic base model is AIO, as it uses a modified text encoder that doesn't make sense to release separately. It's less flexible than the base Anima model in terms of background and environment variety; it tends toward simple, white backgrounds and environments, but that doesn't make it weak.

    The model requires some clarification when generating a character. It is not enough to simply write the character's name and where he is from. I would advise making clarifications, for example, writing his hair color and eye color, perhaps characteristic features, so that the model understands what is required of it.


    The model uses an expanded, previously tested dataset and many of my developments related to the creation of dynamic and living art.


    Workflow for AIO models Anima + lora + upscaler + resizer - YUNI UWU AIO | Anima Workflows | Civitai


    Settings


    CFG 4-6

    Sampler res2s\res_2m\res_3m\ euler\ dpmpp_3m (I tested the model only on them)

    sheduler simple\karras (I tested the model only on them)

    steps to run -1

    steps 10-40 (sometimes you can lower it to 8-9 steps)

    Yes, the model works great at small steps without being a turbo or distillate (those who have used my fine-tuning of ilustrios models know what I mean)

    768x1280 resolution and inversion (I just didn't try others in my tests)

    Description

    This was meant to be a step forward, but it turned out more like a running leap.

    FAQ

    Checkpoint
    Anima

    Details

    Downloads
    142
    Platform
    CivitAI
    Platform Status
    Available
    Created
    10/18/2026
    Updated
    11/17/2026
    Deleted
    -

    Files

    ANIma UWU aio V2 .safetensors

    Mirrors