CivArchive
    Mystique - Krea 2 - Mystique-V1
    NSFW
    Preview 138274590
    Preview 138274698
    Preview 138274639
    Preview 138274500
    Preview 138274501
    Preview 138274502
    Preview 138274552
    Preview 138274556
    Preview 138274562
    Preview 138274563
    Preview 138274564
    Preview 138274570
    Preview 138274609
    Preview 138274611
    Preview 138274686
    Preview 138274719
    Preview 138274735

    Mystique

    This LoRA recreates my interpretation of Mystique as a striking, cunning, and sensual blue-skinned mutant with textured blue skin, long flowing red hair, glowing yellow eyes, and a distinctive small white skull on her forehead.

    Use the trigger word: xmystique

    The LoRA is designed to preserve her recognizable appearance across portraits, full-body images, cinematic scenes, action shots, intimate moments, and a wide range of visual styles. It can produce anything from refined beauty portraits and fashion editorials to dramatic rooftop scenes, rainy night atmospheres, sultry indoor settings, comic-book action, quiet contemplative moments, and more explicit scenes.

    Her default appearance includes the blue textured skin, long red hair, glowing yellow eyes, and the small white skull on her forehead, so these details usually do not need to be repeated in the prompt unless you want to reinforce or slightly alter them. The model also strongly recognizes her signature outfit: the "classic-dress" (a white high-collar sleeveless dress with sheer fabric, an open center with cleavage, and a belt with metallic skulls), often paired with white gloves and white boots.

    She works well in many different settings and styles (photorealistic, cinematic, comic book, graphic novel, 1940s noir, anime, western animation, manga, and more) while also allowing different clothing, casual outfits, or full nudity when desired. The LoRA responds especially well to clear descriptions of pose, body language, expression (sultry, flirty, cunning), camera angle, and lighting.

    The model is most reliable when prompts clearly describe the composition, camera angle, pose, expression, environment, lighting, and any intentional changes to her default appearance or outfit.

    This is a fan creation made out of love for the character and is not affiliated with or endorsed by Marvel. I hope people have fun with it and post your results on the page, I would love to see what you create! Cheers :)

    Description

    FAQ

    Comments (10)

    TheEventHorizonJul 30, 2026
    CivitAI

    Ok she is awesome looking.

    If only krea could do a half way morph effect.

    Dralba
    Author
    Jul 31, 2026· 1 reaction

    Yes, that would have been great! I have considered making a concept LoRA for that type of transformation, but I need to experiment a little bit more :)

    RudyBagaJul 31, 2026
    CivitAI

    This LoRA is not well trained. Using the trigger does not reliably produce the character at all. I'm guessing they captioned with words describing how Mystique looks, because all you reliably get is something kind of like the dress in the example images and an (admittedly) very beautiful black woman.

    Dralba
    Author
    Jul 31, 2026

    That was surprising to hear; it has been working well for me in my generations. Are you sure that the LoRA is activated correctly? I will try it out more, but I haven't encountered the problem. Thank you for letting me know

    TheEventHorizonJul 31, 2026· 1 reaction

    It works fine for me. Try putting in mystique, blue skin, red hair, yellow eyes, or if you want scales, into the prompt. I'll posting something later,

    RudyBagaJul 31, 2026

    @TheEventHorizon Having to prompt for all those things means that the dataset was poorly captioned. The whole point of a LoRA is not needing to prompt for things inherent to the character. This means the creator almost certainly included those things in their captions, which is the opposite of best practice.

    TheEventHorizonJul 31, 2026

    @RudyBaga EH, that's not how krea loras generally work. Usually you need some kind of prompting. I guess a trigger word would help in the future.

    RudyBagaJul 31, 2026

    @TheEventHorizon A single trigger is usually helpful, yes, especially for a character. But you should not be putting a description of the character in the caption when you train. That dilutes the concept.

    Dralba
    Author
    Aug 2, 2026

    @RudyBaga Now I understand what you are referring to. And yes, you are perfectly right, however, using a single trigger also implies that the subject remain similar across the dataset, and I intentionally embedded variations to outfit and some body features, because that is how I like to use the LoRA, and variations must be captured to some extent. There is a tradeoff between a robust trigger word and flexibility. I was aiming for consistent facial and body identity while allowing different outfits and artistic style.

    I can do differently in a future update if you like, if there's a demand, it is an easy fix.

    RudyBagaAug 2, 2026

    @Dralba Yes there are certain things you want to tag if you want them not to appear in all generations, but it appears you must have tagged things that we do want to appear in all generations, like her blue skin. Meanwhile, it appears you did not tag for the white dress, because that is the thing that appears most consistently when using the LoRA.

    Honestly, I was only using it for an experiment with combining LoRAs, I don't need a new version, but thanks for the offer.

    LORA
    Krea 2

    Details

    Downloads
    200
    Platform
    CivitAI
    Platform Status
    Available
    Created
    7/30/2026
    Updated
    8/24/2026
    Deleted
    -

    Files

    Mystique_Krea2_V1.safetensors

    Mirrors