CivArchive
    Kontext Image enlargement without distortion - v1.0

    This workflow uses kontext dev together with Deblur LoRA; the prompts in the workflow maximize consistency of a character’s appearance, but because the control can be too strong it often fails to high‑res the image. In that case you need to adjust the “Deblur lora strength” and “Flux Guidance” values, and you may even need to increase the CFG value.

    Note: this workflow cannot restore old, damaged photos.

    Description

    FAQ

    Comments (3)

    countlippeAug 23, 2025
    CivitAI

    What are your settings in the previews to achieve the results you're getting? My results are turning out quite different to the original photos (gender swaps, non photographic quality etc)

    BBS_revenger
    Author
    Aug 23, 2025· 1 reaction

    If your result deviates too much from the original image, you can try to reduce the strenth of lora or even turn off the "deblur lora". If it still doesn't work, you can try to reduce the number of sampling steps.

    LazmanSep 12, 2025· 1 reaction

    Gender swap(if trying to produce male, especially slender male), is the fault of Flux/SD in general. The original devs that released the base models that everything else is built on, were perverted straight dudes, so like, 90+% of the training was on human females. If they'd been more scientific and kept it in their pants, they would have made a balanced dataset.

    But, thx to them, the most(if not all) the base models, trained on millions(if not billions) of images(a feat. that only the wealthy can pull off due to grfx card costs by Nvidia), think that 'human' = female.

    I've been wrestling with SDXL for ages, and am just finally getting half decent results now. Flux will be worse, cuz most people could never posses the hardware to properly train loras for it, nevermind training/fine-tuning the base models.

    Alls I can say is, add 'male focus(at least once, maybe twice) in your main prompt, make sure to add 1boy(even if your image is of a man, cuz for some reason, boy=man in these things), and under negatives, add '1girl, female, woman, women, tits, breasts, boobs, vagina', and if your character is shirtless, 'nipples, puffy nipples', otherwise it will give your male character triple N (nasty nursing nipples)..

    Additionally, ensure to choose the right model, try to aim for more male focused, or at least neutral models (if there's 18 samples with women, and two/none with men, avoid). The model itself can make a at least a bit of a diff, cuz the wrong ones will absolutely twist and warp your guy into a girl. Other notable tags for negative "panties, pantiehose, pony tail, pig tails, long hair, makeup, lipstick, futa, futanari', etc. and avoid using terms like 'Anatomy', and 'Beautiful' in the positives, as both are largely tied to the female form.

    And if you're trying for a scene without people in it at all, 'no humans' in the positives, may help, as it will typically try to force women into the scene, even if you add all the negs for them, and are prompting no people in the positives.

    Workflows
    Flux.1 Kontext

    Details

    Downloads
    707
    Platform
    CivitAI
    Platform Status
    Available
    Created
    8/21/2025
    Updated
    6/5/2026
    Deleted
    -

    Files

    kontextImage_v10.zip

    Mirrors