CivArchive
    Flux.3 Klein - Klein
    NSFW
    Preview 145782224
    Preview 145782225
    Preview 145782227
    Preview 145782228

    FLUX 3 Image Klein is a 5.6B open weights image model from Black Forest Labs. One model handles text-to-image, editing with up to 10 reference images, and layout control with bounding boxes, at resolutions up to 4K.

    Originally released by Black Forest Labs on Hugging Face. All credit for the model goes to the Black Forest Labs team. Civitai hosts a mirror so creators can generate with it on-site and train against it. Head to the original repo for the reference code and updates.

    Two versions

    • Klein is the distilled model: 16 steps with no guidance, about 6x faster than Base. BFL recommends it for generation and editing.

    • Klein Base is the undistilled model: 50 steps at CFG 3.5. BFL recommends it for fine-tuning, LoRA training and research. LoRAs trained on Base can also be used on Klein, with partial compatibility.

    Files

    Each version carries the diffusion model in bf16, int8 and nvfp4, from BFL's ComfyUI release. The int8 files run on an 8 GB GPU (RTX 3070 class), and nvfp4 needs a Blackwell GPU. The text encoder and VAE ship in the ComfyUI repo.

    Generating on Civitai

    Both versions are in the Civitai generator for text-to-image and editing, with up to 10 reference images. Klein runs at fixed settings. Klein Base exposes steps, CFG and a negative prompt. Bounding-box layout is not in the generator yet.

    License

    FLUX 3 Image Klein is released under the FLUX Kommunity License. The model and anything derived from it, including LoRAs and fine-tunes, are for non-commercial use, and derivatives must be shared under the same license. Users and companies with under US$5M in annual revenue may use the images they generate commercially. Larger commercial use needs a license from Black Forest Labs.

    This FLUX Model is licensed by Black Forest Labs Inc. under the FLUX Kommunity License. Copyright Black Forest Labs Inc.

    Links

    Description

    FAQ

    Comments (17)

    FatbunsOct 15, 2026· 10 reactions
    CivitAI

    Just stick to Krea2. Use whatever for editing. This is not worth it.

    AeralynaiaOct 16, 2026· 2 reactions
    CivitAI

    I can't wait to try this !!! thank you so much !!!!

    ShadowCellOct 16, 2026· 4 reactions
    CivitAI

    https://civitai.red/images/145793589 https://civitai.red/images/145793591

    Initial impressions definitely disappoint (it has a tendency to output a lot of surprisingly bad garbage; bad anatomy, 'Flux face', etc.....). Things that virtually never happen any more in Qwen 2.1 or KREA 2 (so it's a little shocking to still see it here).

    I think this might have big potential in custom models, however.

    It's capable of generating some very high quality images, but you really have to aggressively wrangle it to do so; because it hates SDXL prompt salad (among other things). It's also important to remember that this is part of a newer generation of image checkpoints; that are now really just expecting you to use accompanying side LLMs to generate/enhance prompts. This is just something people are going to have to get used to ('git gud', if you will), because it seems all the image checkpoints are now going this direction (Qwen Image 2.1, Ming, Klein 3). It's looking like it's not going away, no matter how much users want to drag their heels over it, & complain about it.

    That said, a local release of KREA 3 is just around the corner (with edit capabilities); from what I'm seeing of the cloud version, there's a strong chance FLUX 3 might be in some serious trouble lol.

    FLUX 3 is also heavily censored; so how DOA it'll end up being is going to be a matter of how easily the community can work around this (lets face it).

    I'd say give it a fair chance (I was seriously disappointed at first, ngl); but it is generating some very nice images/realistic edits if you persevere.

    ShadowCellOct 16, 2026· 2 reactions

    TL;DR version - This is just an honest appraisal (so people don't waste their time, or end up massively disappointed)..... but if you're going to download this & just 'SDXL prompt salad' away, regardless, then I'd honestly say not to bother.

    mphobbitOct 16, 2026

    You did wrong something https://civitai.red/posts/31677512

    Maybe it's from your LLM encoder (it was too h-ny).

    GodGooner69Oct 16, 2026· 7 reactions
    CivitAI

    Flux 1 was 10x better. No cap. This release rivaled SD3 botched release.

    HzoidOct 16, 2026

    Halloween came early unless there is some bug in the matrix

    mphobbitOct 16, 2026

    It looks like they had problems from large amount of that-stuff loras and checkpoints on FLUX-1 engine.

    HzoidOct 16, 2026· 3 reactions
    CivitAI

    Disclamer for this model you may get sick

    ShadowCellOct 16, 2026· 7 reactions
    CivitAI

    Stable Diffusion 3 FLUX 3 https://civitai.red/images/145794687

    ShadowCellOct 16, 2026

    I joke lol.

    Here's an actually pretty decent image for counter-balance (given I was deliberately trying to break it for a cheap laugh on purpose lol). https://civitai.red/images/145794802

    PrimaveriOct 16, 2026

    Actually the distilled version is very usable! But the base one is just like seing Stable Diffusion comming out from the ashes.

    mphobbitOct 16, 2026

    50-50
    https://civitai.red/images/145806498

    tastyaiartOct 16, 2026
    CivitAI

    WORST NEW MODEL LAUNCH!!

    Wrong anatomy everytime!!!

    ShadowCellOct 16, 2026

    Nah, it's definitely not worse than Pony V7, SD3, or Klein 4B (which was borderline completely useless lol).

    Pony V7 was literally built on the backbone of furry gooners, & then funded for basically just one purpose....... & it then released censored!

    I'm not even a furry in the slightest, & I just visited the Pony V7 page solely to watch the ensuing dumpster fire it was so bad lol.

    mphobbitOct 16, 2026
    CivitAI

    Advanced Flux Klein 4B. Definetely not for generation.

    + It can generate something even pretty.
    + It can into something non-English, even better than Klein 4B. Additional points.


    - Looks like it can't handle long prompts like this: https://civitai.red/images/137343670, gave mess, not published.
    - Low face variety. Looks like it finetuned towards Europeans more than others: https://civitai.red/images/145805759
    https://civitai.red/images/145805763
    - Just look at a poor blonde female on this picture: https://civitai.red/images/145805762
    May be I don't understand something and one knows such real people.
    - Just look at this poor Chinese female: https://civitai.red/images/145805764
    Bad side of SDXL and SD1.5
    - Poor laying woman on a grass: 50-50.
    Simple prompts handles, but agree, advanced prompting turns her into mess of bodyparts :'(
    - I know that people can look double elder than their real age: https://civitai.red/images/145805770

    I didn't tried photo editing but according my Klein 4B experience it has to be much better.

    DanyFitzOct 16, 2026
    CivitAI

    Es muy malo este modelo, a diferencia de krea 2 o qwen imagen 2.1, la primer foto tenia el cuerpo deforme, dedos deformes y un ojo sin terminar de hacer, fue un contraste muy extremo, pasar de hacer algo bien a algo desastroso.

    Checkpoint
    Flux.3 Klein
    by CivitaiOfficial
    Download (Beta) View on CivitAI

    Details

    Downloads
    424
    Platform
    CivitAI
    Platform Status
    Available
    Created
    10/16/2026
    Updated
    10/17/2026
    Deleted
    -

    Files

    flux3Klein_klein_txt_3299383.safetensors

    Size:
    4.52 GB
    SHA256:
    32de113cbf8282b02cbff0794b637b9d6676da09314a1db5a30bcea9bf4ee01c

    Mirrors

    HuggingFace (4 mirrors)
    CivitAI (2 mirrors)
    ModelScope (1 mirrors)

    flux3Klein_klein_pruned_int8.safetensors

    Size:
    5.43 GB
    SHA256:
    847d8933548afd93df4af2cac2498e70891bf28d9dae06d695f536992e168fe2

    Mirrors

    HuggingFace (3 mirrors)
    CivitAI (2 mirrors)
    ModelScope (1 mirrors)

    flux3Klein_klein_txt_3299287.safetensors

    Size:
    9.08 GB
    SHA256:
    cb8cb9b9dcd2569509de755dca54d96fd93a16a7aabfd2f4051f0a144230620f

    Mirrors

    HuggingFace (2 mirrors)
    CivitAI (2 mirrors)
    ModelScope (1 mirrors)

    flux3Klein_klein_txt_bf16.safetensors

    Size:
    8.27 GB
    SHA256:
    f0b24fc5c8d7c0ea6ca909da14c20af5cac53f22506be942c6c956f838b9630f

    Mirrors

    HuggingFace (5 mirrors)
    CivitAI (2 mirrors)
    ModelScope (2 mirrors)

    flux3Klein_klein_bf16.safetensors

    Size:
    10.34 GB
    SHA256:
    c3e9813cdd049c831c836b2e9db61150e49a93fc1a4c63b256494642fee0bd7c

    Mirrors

    HuggingFace (7 mirrors)
    CivitAI (1 mirrors)
    ModelScope (2 mirrors)

    flux3Klein_klein_nvfp4.safetensors

    Size:
    4.92 GB
    SHA256:
    15d421e2558443fcf83ba46f3cd884d9d181c2c7a0a7bdcd2974af234a2d4f5c

    Mirrors

    HuggingFace (2 mirrors)
    CivitAI (1 mirrors)
    ModelScope (1 mirrors)

    flux2-vae.safetensors

    Size:
    320.64 MB
    SHA256:
    868fe7b343cc8f3a19dbcfcafbc3d5f888802be3f89bd81b65b3621a066ce8f3

    Mirrors

    HuggingFace (136 mirrors)
    CivitAI (10 mirrors)
    ModelScope (5 mirrors)
    ModelScope CN (4 mirrors)