CivArchive
    YumeWa Real - v1.0
    NSFW
    Preview 131629511
    Preview 131629546
    Preview 131629578
    Preview 131629585
    Preview 131629630
    Preview 131629653
    Preview 131629668
    Preview 131629679
    Preview 131629686
    Preview 131629689
    Preview 131629703
    Preview 131629708
    Preview 131629717
    Preview 131629721
    Preview 131629724
    Preview 131629746
    Preview 131629774
    Preview 131629887
    Preview 131629769
    Preview 131629919

    After careful fine-tuning, I would like to release the YumeWa Real Illustrious checkpoint. First, let me introduce the predecessors:

    Base Model: KodoraIL 17.2 - by Kodora
    This is an awesome model that takes realism and stretches it to the boundaries of fantasy. It does great with different angles and body styles, creating the flexibility that I wanted to have as the core of my models, something that is often overlooked but so nice to have in realism models. Thank you Kodora for allowing merges.

    BeretMix Real v10 - by AI_beret - a long standing favorite for Japanese faces and realism, this model also has a lot of flexibility and a beautiful Japanese face.

    Note that Kodora's restrictions have been applied to this model.

    YumeWa Real

    Yume is Japanese for Dream and Wa mean Harmony. I chose these names to represent this model because the subject matter is broad and you can dream up whatever you like, and the merge is harmonious and does not falter easily, leading to generations that accept the tags in your prompt without much work.

    What YumeWA Gives You

    Faces that lean towards the BeretMix appearance, with tons of flexibility and the ability to explore concepts that other realism checkpoints choke on.

    Some bias towards NSFW images.

    Compatibility with most LoRAs: detailers and lighting LoRAs such as Envy, Dramatic Lighting Slider, and more

    T-Girl (futa) capable but not the strength of this model. Prompt carefully and try different seeds and prompts/weights if at first you don't succeed.

    Gallery:

    All images in the checkpoint gallery were generated without upscaling, with exception of the image with 4 women.
    All images have been generated using at or near the following:
    Sampler: Euler A

    Steps: 22

    CFG: 5.2-5.8

    However you may experiment with other settings freely.



    I noticed for fine details there was a bit of residual noise I was unable to tweak out when lowering the steps to 20, but in general this is not a problem and only demonstrates that the model is working exceptionally well, with no blotchiness, color artifacts or other major defects.

    Feel free to experiment with other sampler/scheduler combinations. Exp_Heun / Exponential is one of my favorites. DPMPP_2M_SDE + Karras is also a very strong combo.

    Description

    Early Access helps me reward artists and build a strong community around the model. I don't keep this buzz, it goes straight back to creators, especially those who use on-site generation.

    FAQ

    Checkpoint
    SDXL 1.0

    Details

    Downloads
    106
    Platform
    CivitAI
    Platform Status
    Available
    Created
    5/25/2026
    Updated
    5/25/2026
    Deleted
    -

    Files

    yumewaReal_v10.safetensors

    Mirrors

    CivitAI (1 mirrors)