CivArchive
    KREA2-PROPER-WF - PERFECTA
    NSFW
    Preview 136897593

    Here is the best KREA2 workflow for busy content creators that just want to produce stunning photos of their AI (slop) virtual influencers ;)

    you will need to download :
    https://civarchive.com/models/2729562/krea-2-real-vae
    https://huggingface.co/ahmed22xa/Huihui-Qwen3-VL-4B-Instruct-abliterated-comfy
    https://huggingface.co/wikeeyang/Krea2-Turbo-HD-V1
    optional:
    https://civarchive.com/models/2728644/krea-2-mystic-xxx

    alternative TE: https://huggingface.co/DreamFast/Qwen3-VL-4b-Heretic-ComfyUI/tree/main or https://huggingface.co/DreamFast/Qwen3-VL-4b-Heretic-GGUF

    alternative VAE: https://civarchive.com/models/2718533/hdr-vae-anima-krea2-qwen-image

    for PiD in PERFECTA get the Qwen-Image version of PiD model + TE: https://huggingface.co/Comfy-Org/PixelDiT

    UPDATE: the best settings for steps/scheduler/sampler is here in this image (download as PNG and drag into your ComfyUI): https://civarchive.com/posts/29987247

    Description

    this version uses Nvidia PiD (High-Resolution Latent Decoding with Pixel Diffusion) for those who are true PERFECTIONISTS!

    FAQ

    Comments (8)

    menorpeca59105Jul 16, 2026
    CivitAI

    I tried to use it, but it showed a black screen 😭

    0l1v1aR0551
    Author
    Jul 16, 2026

    maybe you need to turn off attention (like SAGE ... etc)

    CowgaryJul 17, 2026

    Tried the following things:

    1. Use KJ loader instead to manually toggle Sage Attention, Sage = on & Krea2T Enhancer Advance = on -> black picture
    2. Use KJ loader Sage = off & Krea2T Enhancer Advance = on -> black picture
    3. Use KJ loader Sage = off & Krea2T Enhancer Advance = off -> picture rendered
    4. Use the original loader in WF & Krea2T Enhancer Advance = off -> picture rendered

    Using RTX 3090 and Qwen VAE.

    0l1v1aR0551
    Author
    Jul 17, 2026

    @Cowgary why are you using qwen vae which is the lowest quality vae ever??

    voboyso632Jul 17, 2026· 1 reaction

    @menorpeca59105 Try setting the Krea2T-Enhancer-Advanced node to strength 1, or disable.

    menorpeca59105Jul 17, 2026· 1 reaction

    @voboyso632 I tested it and it worked, but the only problem is that without it turned on, it ends up ignoring half of the prompt.

    CyclopsGERJul 16, 2026
    CivitAI

    Hi, I loaded your workflow and all required models but I am getting:
    "KeyError: 'int8_tensorwise'" on Load Diffusion Model / Krea2-Turbo-HD-V1-int8_convrot.safetensors

    0l1v1aR0551
    Author
    Jul 16, 2026

    update comfy to the latest!

    Workflows
    Krea 2

    Details

    Downloads
    130
    Platform
    CivitAI
    Platform Status
    Available
    Created
    7/16/2026
    Updated
    9/19/2026
    Deleted
    -

    Files

    krea2PROPERWF_perfecta.zip

    Mirrors

    CivitAI (1 mirrors)