CivArchive
    Kreanima - v1.0
    NSFW
    Preview 137644828
    Preview 137644832
    Preview 137644784
    Preview 137644778
    Preview 137644789
    Preview 137644811
    Preview 137644802
    Preview 137644809
    Preview 137644798
    Preview 137644813
    Preview 137644817
    Preview 137644819
    Preview 137644821
    Preview 137644822
    Preview 137644824
    Preview 137644835
    Preview 137644836
    Preview 137644837
    Preview 137644990
    Preview 137644841

    This is a bit of a broken merge of Krea2, based on my own LoRA, provided as-is, published here only to link my own local experiments with Krea2.

    TL; DR: Use at your own risk 😅 (mainly sanity and time are what you could lose🤣).

    It works with stable-diffusion.cpp and should work with other tools that accept GGUF files.

    Recipe

    It is leveraging a mix of DMerge (marked as allowing merge at time of integration) and the Vanilla version of Krea2 Turbo plus a 50/50 mix of the uncensored and vanilla version of Lianima 4 Krea, at 0.8.

    Due to the inclusion of uncensoring stuff and my own OC, the result is a highly unruly, very NSFW, woman focused, anime version of Krea2 in my own style 😘

    No trigger word needed, use as you would use Turbo (CFG ~1, Steps ~8)

    • LoRA merge was done painfully manually until i can find a way to do it faster/easier with sd-mecha (unlike ANIMA, i can't hold the whole BF16/FP16 version in memory).

    • Quantization was done thanks to the new "stream layers" feature of sd.cpp:

    sd-cli.exe -M convert -m kreanima_fp16.safetensors -o kreanima_Q5_0.gguf --type q5_0 --stream-layers --max-vram -1 --offload-to-cpu

    PS: It took 9 hours but i have also uploaded the FP16 version. I will not provide any other variant.

    Thanks for reading! 💜

    For reference, prompt for the picture at the top:

    [character description]
    1girl, Liadora, red hair, green eyes, black oni horns, black dress, purple elbow gloves, upper body, shy smile
    [scene description]
    Liadora is on the left side of the screen, holding a board with "Kreanima" written on it, showing to the viewer. Abstract background, geometric patterns, neon colors.

    Description

    Posted on Civitai, upload found elsewhere are FORBIDDEN and done without the author permission.

    FAQ

    Comments (16)

    pezpok1929Jul 24, 2026
    CivitAI

    How do i get this running on comfyui? I've tried in the unet folder to load the GGUF node.

    n_Arno
    Author
    Jul 24, 2026· 1 reaction

    I don't use comfy and can't help :-(

    pezpok1929Jul 24, 2026

    Its all good. I shall wait to see if you release others. all good :D

    oneforbackupJul 24, 2026

    I use Molbal's Comfyui-gguf custom nodes for KREA2, I don't know if the other gguf nodes are patched for Krea2 yet. However I have not downloaded this model to test if it works.

    physisJul 24, 2026

    yeah it doesn't work on comfyui I think

    pezpok1929Jul 24, 2026

    @oneforbackup thanks i will give it a go some time thanks :)

    n_Arno
    Author
    Jul 24, 2026

    @pezpok1929 i will try "once" to upload the full 24GB model, but when an upload fail with Civitai, you need to redo it from scratch T_T (and it will probably takes several hours to do so)

    pezpok1929Jul 24, 2026· 1 reaction

    @n_Arno dont stress. Its all good. If you cant, its fine. Im not one of those to get angry if I cant get a checkpoint.

    cl4usaJul 24, 2026· 2 reactions
    CivitAI

    Your sample images look great! Do you have any plans to release a INT8 or FP8 version?

    n_Arno
    Author
    Jul 24, 2026

    As i don't use comfy, i don't think i'll release a int8 convrot or fp8 scaled version (i didn't check how those are made or works).

    I'll "try" releasing the fp16 (i made a mistake at one point and lost the bf16 quantization), but it will be a gamble on how the 25GB upload goes... 😔

    aa9739614Jul 25, 2026· 1 reaction
    CivitAI

    各位,5060ti 16G是不是不能跑,看模型24G了

    n_Arno
    Author
    Jul 25, 2026

    I can make the model run on a RTX 3060 with 12GB of RAM, but i use the GGUF quantized version. An INT8 Convrot version working for ComfyUI could be a solution, but i need to check https://github.com/BobJohnson24/ComfyUI-INT8-Fast to see how and if i can make it.

    pezpok1929Jul 26, 2026· 2 reactions
    CivitAI

    Great checkpoint. thanks heaps :)

    n_Arno
    Author
    Jul 26, 2026
    CivitAI

    Thanks @GombaTheGreat for the tips! 💜

    n_Arno
    Author
    Jul 29, 2026
    CivitAI

    Thanks @Kal350 for the tips! 💙

    GuideboAug 1, 2026
    CivitAI

    Hi, thanks for the checkpoint. I tried to use the GGUF Q5 version (with the latest version of Forge Neo), but it doesn't work, it doesn't even start generating and I get all kinds of errors (never had this with other GGUF checkpoints, although they're Q4) and the fp16 version is just to big for my GPU. I really hope you could also release an int8_convrot version, or even an fp16 in half the size or fp8. I really like the style of this checkpoint.

    Checkpoint
    Krea 2

    Details

    Downloads
    352
    Platform
    CivitAI
    Platform Status
    Available
    Created
    7/24/2026
    Updated
    8/4/2026
    Deleted
    -

    Files

    kreanima_v10.gguf

    Mirrors

    HuggingFace (1 mirrors)
    CivitAI (1 mirrors)

    kreanima_v10.safetensors

    Mirrors

    HuggingFace (1 mirrors)
    CivitAI (1 mirrors)