CivArchive
    realisticnoobmix - v2.0
    NSFW
    Preview 60815137
    Preview 60826604
    Preview 60945316
    Preview 60945505
    Preview 60945702

    The new version is more realistic, but its level of realism is not as high as that of a true realism model. This is because, during the model fusion process, I discovered that the layer controlling the realism of the image is the same layer that controls the character's body shape. To ensure that the characters maintain body proportions similar to those of anime characters, I had to sacrifice a bit of realism

    just a merge of Willy's Noob Realism and CyberIllustrious | CyberRealistic

    my goal is to get a realistic checkpoint that is more friendly to noobai lora

    and I'm happy with the result

    Description

    merge of homosimileXL_v30NAIXL, quintillusIllustrated_v2CloserToReality, cyberillustrious_v30

    FAQ

    Comments (4)

    BLACKbaiMar 7, 2025
    CivitAI
    Your base model is great, I love it and I love you too


    LazmanMar 21, 2025
    CivitAI

    Since you're savvy in the art of mixing individual layers, have you attempted to merge realism layers from SDXL 0.9 with 1.0?

    Cuz with 1.0, they gutted the model so badly to censor it and reduce size, that they removed SDXL's ability to create truly photo-realistic images with actual shading and other elements that made the image actually look real.

    The original SDXL0.9 is 12-13gb in size, but someone did make a pruned one that keeps the perks of 0.9 but around the size of 1.0. So, that one may be more compatible for merges?

    Or otherwise, if you can simply merge individual layers from 0.9 with 1.0 then perhaps you could pull realism layers from the original 0.9 to merge with 1.0, but without the layer conflict you spoke of in the description?

    I'm a bit of a newb in terms of any form of full model training and such, (cuz I don't got a million dollar PC), so I'm just theorizing here. But, if it were possible to reclaim the realism of the 0.9 model, then you could really get a jump on most other SDXL-based realism models.

    mingliu169409
    Author
    Mar 21, 2025

    Thank you for your advice. My goal in merging models is to adapt to the LoRA trained by noobai. Based on my attempts, if I use the SDXL model for merging, a low ratio doesn't yield effective results, while a slightly higher ratio causes the loss of character features from the LoRA. Additionally, the hand rendering in SDXL is bad.

    LazmanMar 23, 2025

    @mingliu169409 I do understand your points, but what of 0.9? Again, I'm a bit of a newb in that area, so that's why I'm asking you, who seems to know a bit more. Like, if you merge realism layers from the original 0.9 or even the prune into 1.0(assuming it's possible), then would it still be the same layers that control the character features?

    I mean, they were able to remove those layers from 0.9 without destroying character features, so..

    Also, I don't know much about noobAI, I've only heard of it. Is it another brand of SDXL like illustrious or a different brand altogether like hunyuan? Also, which lora do you mean, or is there only one lora for it?

    Checkpoint
    Illustrious

    Details

    Downloads
    426
    Platform
    CivitAI
    Platform Status
    Available
    Created
    3/1/2025
    Updated
    5/12/2026
    Deleted
    -

    Files

    realisticnoobmix_v20.safetensors

    Mirrors