CivArchive
    Qwen-Image-Edit FP8 & GGUF - VAE
    NSFW

    Error: Can't find mmproj file for 'Qwen2.5-VL-7B-Instruct-Q3_K_M.gguf' (matching:'qwen2.5-vl-7b-instruct')! Qwen-Image-Edit will be broken!

    Rename the mmproj-F16:

    Make sure the downloaded text encoder has the correct file naming (whatever_text-encoder.gguf paired with whatever_text-encoder_mmproj-F16.gguf)

    Demo: https://chat.qwen.ai/
    Text encooder FP8: https://civarchive.com/models/1864281?modelVersionId=2110105

    Workflow GGUF: https://civarchive.com/models/1881851/qwen-iamge-edit-gguf-lightning-4steps-comfyui

    This FP8-optimized (e4m3fn) conversion of Qwen-Image-Edit, now fully compatible with the latest Qwen-VL image editing features! Leveraging Alibaba's advanced multimodal AI, this model enables seamless text-guided image manipulation, perfect for precise edits, prompt-based adjustments, and creative refinements. The model is mirrored here for convenience.

    Description

    FAQ

    Comments (6)

    instabackpackers999Aug 21, 2025· 1 reaction
    CivitAI

    am started getting black image in output. am using gguf

    5817430Aug 21, 2025

    I believe it's because of Sage Attention. what tool do u use? ComfyUI? portable or github version? ask Grok or Gemini about how to start your tool without it,

    expert78Aug 22, 2025
    CivitAI

    Does fp8 model fit in 16 gb vram

    5817430Aug 22, 2025

    easily, you can run full distilled FP8 without any problems, much better than regular FP8

    SkoomaCatJesusAug 24, 2025
    CivitAI

    Is the clip model the same for image and edit?

    5817430Aug 24, 2025

    [Uploaded the mmproj-f16 related to the 'instruct' text encoder. see the screenshot in the description] I tested the Edit model with the Qwen-Image text encoder when it was released and observed a different behavior from GGUF 'instruct.' I'm not sure because right now the ComfyUI implementations for Qwen-Image-Edit are not bug-free, especially when it comes to GGUF. The size of the image overrides the prompt, and normally you would think something is wrong with the text encoder, CFG, or steps. unsupported image sizes are something you need to pay attention to, even if the text encoder model is correct.

    Checkpoint
    Qwen
    by civitai
    Download (Beta) View on CivitAI
    qwen
    qwen edit
    base model
    qwen image

    Details

    Downloads
    1,073
    Platform
    CivitAI
    Platform Status
    Available
    Created
    8/19/2025
    Updated
    8/17/2026
    Deleted
    -

    Files

    qwenImageEditFP8GGUF_vae.safetensors

    Size:
    242.05 MB
    SHA256:
    a70580f0213e67967ee9c95f05bb400e8fb08307e017a924bf3441223e023d1f

    Mirrors

    HuggingFace (229 mirrors)
    CivitAI (175 mirrors)
    TensorFiles (1 mirrors)
    ModelScope (3 mirrors)
    ModelScope CN (5 mirrors)