CivArchive
    [LuisaP❤️] QWEN-EDIT UPSCALER & DENOISE V3 FIXED - v2.0
    NSFW
    Preview 108561946
    Preview 108561939
    Preview 108562484

    V3: Solve the saturation problem + has more variated dataset, ill do v4 with more sharpen for images that already are bigger

    Made with collaboration with @artificialguybr

    this actually upscale compressed and bad looking images, but dont work on blurred image and very pixelated ones.

    to upscale more than 1024x1024 as default of qwen, you need to change your workflow to upscale the image in latent before using the lora.

    Prompt: upscale this image

    Description

    FAQ

    Comments (12)

    Error666Nov 3, 2025· 2 reactions
    CivitAI

    I use v1 and v2 simultaneously, it's simply incredible. For restoring a low-resolution photo, even with paid tools, I don't think I've ever been so satisfied with the result. The rendering is very natural.
    Thank you very much.

    If the result is too saturated or lacks naturalness, slightly lowering the denoise seems like a good idea. Or lower the Lightning LoRa and add steps.

    cranpeachNov 3, 2025· 3 reactions
    CivitAI

    Love the model. Going to stick with 1.0 for now as my workflow with 2.0 is causing it to become way too saturated. I read the other comment with tips but if doing a v3 would appreciate if the saturation issue is fixed.

    JellaiNov 3, 2025· 1 reaction

    Yeah, I am really impressed with v2, but the standard saturation addition is too much. Maybe that should be a separate part of the prompt? Some photos are just compressed and low res, and their color data is fine.

    luisa_pinguin
    Author
    Nov 4, 2025· 4 reactions

    solved

    cranpeachNov 4, 2025· 1 reaction

    @luisa_pinguin i appreciate your quick work, unfortunately this one isnt working at all for me even at 2.0 strength. Result spits out the image unchanged. Might just be my workflow/image choice but tried v1 on same image and it worked. Is the 15mb file size intentional? Havent seen Loras that size since SD 1.5 lol

    luisa_pinguin
    Author
    Nov 4, 2025· 1 reaction

    @cranpeach i've trained a lokr with auto mode, i'm trying different stuffs

    JellaiNov 4, 2025· 1 reaction

    @luisa_pinguin The v3 lora is only 15MB. It seems like an error or mistake. I'm not sure even SD1.5 loras are that small.

    luisa_pinguin
    Author
    Nov 4, 2025· 2 reactions

    @Jellai so i'll delete the v3 and try again

    luisa_pinguin
    Author
    Nov 5, 2025· 2 reactions

    did a fix, can you try it again?

    cranpeachNov 5, 2025· 1 reaction

    @luisa_pinguin Very impressive so far!

    cranpeachNov 5, 2025· 1 reaction

    @luisa_pinguin This might be the best upscaler I've used. You can hide this comment if you'd like, I'll leave a new one lmao

    arjunharjunh736Nov 12, 2025

    share the workflow please

    LORA
    Qwen

    Details

    Downloads
    524
    Platform
    CivitAI
    Platform Status
    Available
    Created
    11/2/2025
    Updated
    5/12/2026
    Deleted
    -
    Trigger Words:
    upscale this image

    Files

    qwen-edit_upscale.safetensors

    Mirrors

    HuggingFace (1 mirrors)