CivArchive
    RedCraft | 红潮 | KREA 2 赤佬 Red Mix Edition NSFW (GGUF) - v1.0
    NSFW
    Preview 135065182
    Preview 135064988
    Preview 135064999
    Preview 135065012
    Preview 135065025
    Preview 135065600
    Preview 135065896
    Preview 135066650
    Preview 135066984

    This is a GGUF conversion of https://civarchive.com/models/958009/redcraft-or-or-krea-2-red-mix-edition-nsfw?modelVersionId=3066243

    Patched ComfyUI-GGUF nodes that support it I'm ComfyUI: https://github.com/molbal/ComfyUI-GGUF

    RedCraft-红潮 NSFW

    做好工具人 服务艺术家

    Forever in memory of METAFILM Studio founder Mr. Yuan Bo

    Krea2-RED-Mix 赤佬版 25/06/2026 10Steps
    Raw+Turbo Merged

    Reveal every detail
    KREA 2 赤佬 Bastard Edition
    Uninhibited & zero censorship

    Usage ：ER_SDE/Euler | Simple | CFG=1 | 10Steps

    Description

    FAQ

    Comments (13)

    TaiLongJun 27, 2026· 1 reaction
    CivitAI

    Thanks a lot mate 👍 💙

    filatkin_mJun 27, 2026
    CivitAI

    The GGUF conversion turned out to be 1.3 GB larger than the safetensors original!

    ss9999Jun 28, 2026

    As far as I know, Q8_0 has lower perplexity than FP8. It would be ideal, of course, to make a Q8_0 GGUF from an FP32 source.

    lolmao500Jun 28, 2026

    Lots of times the Q8 version is better than the FP8 version. On AI video models at least, dont know for sure on image models.

    wangf35979Jun 28, 2026

    I'm also curious about the file size increase issue. Of course, thank you for your work!

    famalamJun 28, 2026

    use the q4 version then - you muppet.

    molbal
    Author
    Jun 28, 2026· 2 reactions

    @wangf35979 All is here, this what I used to convert: https://github.com/molbal/ComfyUI-GGUF/blob/main/tools/convert.py - I had to add this to work with Redcraft https://github.com/molbal/ComfyUI-GGUF/commit/73439d2120c2e465fbdf7eb9e98c17ef592e5da3 I also had to re-quantize from FP8 (I converted Krea 2 Turbo and Raw from FP or BF16)

    12ghjjJul 2, 2026· 1 reaction
    CivitAI

    bless you...

    ParadiseLABJul 12, 2026
    CivitAI

    what VAE and Text Encoder should I use?

    molbal
    Author
    Jul 12, 2026· 1 reaction

    I recommend WAN 2.1 VAE, and Qwen3 VL 4B Instruct Text Encoder.

    ParadiseLABJul 13, 2026

    @molbal ok thx

    jeremypierre076910Jul 16, 2026
    CivitAI

    Would be nice to see the Q8 vs Q4 comparisons without toasting my old card in trials.

    molbal
    Author
    Jul 16, 2026· 1 reaction

    Agreed. I did not see a noticeable difference between Q4 and Q8, uploaded images are rather random. However I also only have a laptop, if I ever get my rtx 3090 working, I will make comparisons.

    Checkpoint
    Krea 2
    by molbal
    Download (Beta) View on CivitAI
    base model

    Details

    Downloads
    1,635
    Platform
    CivitAI
    Platform Status
    Available
    Created
    6/27/2026
    Updated
    8/21/2026
    Deleted
    -

    Files

    redcraftKREA2RedMix_v10.gguf

    Size:
    7.74 GB
    SHA256:
    3c5dd604d63884d7146b5242ad5a97933d4eac608e4571e9a4827e4f701a2566

    Mirrors

    CivitAI (2 mirrors)

    redcraftKREA2RedMix_v10.gguf

    Size:
    13.56 GB
    SHA256:
    96f33d068f0e3323288fd006b9c8799a794e9159fabe26328b661d5a40fa24e6

    Mirrors

    HuggingFace (8 mirrors)
    CivitAI (1 mirrors)