CivArchive
    SNOFS (Sex, Nudes, and Other Fun Stuff) Flux 2 Klein 9b Base & Distilled - v1.4 Base
    NSFW
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined

    These are simply a merged in version on top of Flux 2 Klein 9b Base or Distilled with a little extra training for those that can't use lokrs for whatever reason. For full details on how best to use this model please check out the lokr page: https://civarchive.com/models/1972981/sex-nudes-other-fun-stuff-snofs

    Both the lokr and this trained model are released under this license (parts of which shouldn't really matter because Flux 2 Klein is under a noncommercial license...):
    https://huggingface.co/Ashen3/SNOFS

    Want to support my work or help fund the training of this dataset on other models? Join the Patreon in my profile, and if you do - thank you!

    Description

    FAQ

    Comments (10)

    sonjasapphire314May 24, 2026· 4 reactions
    CivitAI

    Unfortunately, this model won't be usable by many users: it's not FP8 and it's not 4-step - people don't have enough RAM, they don't have enough VRAM, they have slow video cards, and connecting LoRA slows down generation by half.

    aising23May 24, 2026· 1 reaction

    V1.2 distilled is FP8, and there is also an BF16 distilled version V1. These can generate in 4 steps out of the box.
    9b Base (V1.0, 1.3 and 1.4) is for training LoRa/LoKr, but it can also generate in 4 steps if you add the Turbo LoRa to the workflow (CFG 1 for fastest generation).
    You are right that VRAM is a bottleneck for many. Offloading is an option for 9b Base, for a decent system, but that is slow.
    Maybe a 4b model will also be released(?)

    sonjasapphire314May 28, 2026

    @aising23 I don't want to use klein 4B. I have GTX 1050 with 4 GB VRAM and still can use klein 9B FP8 distilled, 768x512 generation takes about 1 minute - it's ok.

    sonjasapphire314May 25, 2026· 2 reactions
    CivitAI

    Please make 4 step turbo versions in FP8, because using turbo LoRA on klein-9b-base slows down generation by half, and converting to FP8 when loading the model doesn't work for some reason. Thank you in advance!!!

    MrToonMay 25, 2026· 3 reactions
    CivitAI

    Love your model but wish you kept on going with the distilled versions. Especially with the omni workflows out there that work with ComfyUI-Flux2Klein-Enhancer its hard to make the base model perform just as good. Currently stuck on 1.2 because 1.4 wont play nice with those nodes.

    Ashen3
    Author
    May 26, 2026· 5 reactions

    I'll get it up sometime soon.

    sonjasapphire314May 26, 2026· 1 reaction

    @Ashen3 distilled and FP8 also, please :)

    Ashen3
    Author
    May 29, 2026· 5 reactions

    1.4 is up.

    MrToonMay 30, 2026

    @Ashen3 you are awesome

    sycotix416938May 28, 2026
    CivitAI

    Is there a recommended workflow for this?