CivArchive
    Qwen 2.1 FinalCut (3.8-6.6GB) 25 Step - v1.0
    NSFW
    Preview 144985564
    Preview 144985560
    Preview 144985561
    Preview 144985563
    Preview 144985567
    Preview 144985566
    Preview 144985569
    Preview 144985571
    Preview 144985572

    Qwen 2.1 FinalCut (6.6GB) 25 Step

    After 20 failed trainings including a training that took 4 A100 80GB a few days I finally have a version I am happy to release.

    • 25 Steps

    • Reduced "QWEN Canvas" issue.

    • Works with most systems without issues and has lora support.

    Note: I settled on group size 32 and standard qwen exclusions for INT8

    I found the single DiT blocks of this model way worse to train then the double/single Krea2 - I personally do not like this model to train, and for use I would rate it above SDXL and Flux2. It handles text placement and guidance very well, but seems to fight new concepts.

    Description

    FAQ

    Comments (9)

    HT83Oct 8, 2026
    CivitAI

    Is it good for image edit?

    Felldude
    Author
    Oct 8, 2026· 1 reaction

    It should still take the multi-image and transparency input, but as that is not part of comfy core I did not test

    ArttTakuOct 8, 2026
    CivitAI

    Looks very good, but I'm gonna wait on users commenting on its editing capabilities before plunging on yet another model.

    Generated_GoodOct 8, 2026· 1 reaction
    CivitAI

    I did a couple of test. It works for image editing. Unfortunately for T2I I had more misformed limbs that with the base mode. what is the recommended scheduler and sampler?
    I can't wait for V2

    does it preserve faces

    Felldude
    Author
    Oct 8, 2026

    Normal/Euler

    infernahermit846Oct 8, 2026· 1 reaction
    CivitAI

    Edit: Int8 added, now we are talking :D

    Thank you for the model.
    It is near 2x slower than Int8 models (3060ti).

    It does edit just fine.
    It generates fine.
    It can use loras i tested just fine (consistency, turbo, BFS head swap)
    For edit, Turbo loras shift the image alot.

    Overall, not for me since it takes twice as much to generate the same quality as base or other models.

    ferrrett33Oct 13, 2026· 1 reaction
    CivitAI

    Bro I have no idea what or how you're training, but I ran my standard NSFW test on it comparing it to the base model, which took about 2 hours and all 50 prompts come out either just as bad or sometimes worse.

    I used CFG 3, Euler Simple, since Qwen with CFG 1 is a disaster, and this test does work.

    Could you tell us more about what this is intended for?

    Felldude
    Author
    Oct 13, 2026

    So when testing I do seed to seed, ranking failures and successes - At cfg 1 I did not have a single passable image at 816x1232 or 1280x1920 for xxx on qwen 2.1 base - I had many many aberations - Granted this finetune did not improve xxx much but for x I had more successes seed to seed

    Checkpoint
    Qwen 2.1

    Details

    Downloads
    3,323
    Platform
    CivitAI
    Platform Status
    Available
    Created
    10/7/2026
    Updated
    11/22/2026
    Deleted
    -

    Files

    qwen_image_21_FinalCut_int8_convrot.safetensors

    qwen_image_21_FinalCut_FP8.safetensors

    Mirrors

    qwen_image_21_FinalCut_NVFP4.safetensors

    qwen_image_21_FinalCut.safetensors

    Mirrors