CivArchive
    Satin String Bikini Panties - V1.0 Illustrious
    NSFW
    Preview 149143634
    Preview 149143635
    Preview 149143636
    Preview 149143637
    Preview 149143638
    Preview 149143639
    Preview 149143641
    Preview 149143643
    Preview 149143646
    Preview 149143649
    Preview 149143653
    Preview 149143661

    Satin String Bikini Panties: I was and still am a huge fan of this style of panties, I wish they would get popularized again. Shiny solid-color satin string bikini panties with the zigzag-stitched elastic trim along the edges and side straps, and slightly loose fitting fabric that has wrinkles in it like real non-stretch satin panties would have. No wide band straps, no lace edges or any other designs in the training data, just old school satin string bikini style.

    Running generated images through a detailer pass on the panties helps dial them in a lot better. The zigzag stitching on the trim can generally turn out looking like horizontal lines on the initial generation. Running a detailer pass usually corrects the edges and straps to have a proper zigzag stitch in my experience. The Krea-2 version seems to handle adding text print to the panties very well depending on the specific finetune used. Sometimes it can handle adding patterns to them, but that is wildly unreliable at the moment. I will be training another version on a much larger dataset that handles different patterns and prints as well once I finish curating it.

    The Illustrious V1 version seemed to hyper-saturate the colors when using the base model, but it works a bit better on finetunes in my experience. The V2 version seems to have fixed the hyper saturation a bit more, and improve overall.

    Trigger: satin_string_bikini_panties, then describe them with a color, e.g. shiny red satin string bikini panties. Trained colors: red, black, white, hot pink, baby pink, purple, baby blue, electric blue, yellow, neon green, neon orange; other colors work too.

    Strength: 0.8 - 1.0.

    Note on art style: This first version was trained entirely on real photos, so it tends to pull the art style toward realism. This is more apparent on models like Illustrious or Anima. You can override the photoreal bias with a stylized checkpoint (for example an anime Krea 2 finetune) plus artist / style LoRAs, which brings it back to anime, 2.5D or illustrated looks while keeping the garment. This works really well with Krea 2, but I haven't tried it much with the other models, though I imagine it likely works similarly. For stylized work, use a lower strength (0.6 - 0.8) and strong style tags. I usually add terms to the prompt like 'glossy fabric', 'zigzag stitching on trim', and 'subtle wrinkles in the fabric' out of habit, but the images seem to come out fine without those terms. Try them out and let me know if you find issues or anything odd, the info might help with the V2 retrains coming next.

    I need more non-photorealistic images to create a dataset that includes additional art styles beyond photoreal, and these type of panties went out of style mostly in the late 90s/early 2000's, so they don't really exist. Hopefully by releasing this to the world, more images generated in different art styles can be created and help me gather enough to curate a dataset containing multiple art styles to prevent the photorealistic bias of V1.0. As I said, using stylized finetunes and artist style loras can get rid of the photorealistic pull.

    Side by side showcase images: the same seed without the LoRA on the left and with it on the right.

    Free to use, merge and share. No early access, no paywall. The models themselves are open source, and they were trained on publicly available images. Seems like a dick move to charge people for things I trained on top of it from additional publicly available images.

    I'm working on making more varied images to add to the showcase image collection. This is the very first LoRa I have ever trained, so I am getting the hang of all this still.

    Description

    Illustrious version, step 1000. File: Panties-CLOTHING-Satin_String_Bikini_Panties-V1.0-Illustrious.safetensors

    Known issue: on the base Illustrious XL model this V1.0 over-saturates colours and softens fine detail compared with the same seed without it (strength 0.6 - 0.8 helps). It behaves much better on Illustrious finetunes such as DaSiWa Sweet Poison and Animosity (see the showcase). A V2 trained on booru-tag captions to fix this is in training.

    FAQ

    LORA
    Illustrious

    Details

    Downloads
    56
    Platform
    CivitAI
    Platform Status
    Available
    Created
    11/15/2026
    Updated
    11/18/2026
    Deleted
    -
    Trigger Words:
    satin_string_bikini_panties

    Files

    Panties-CLOTHING-Satin_String_Bikini_Panties-V1.0-Illustrious.safetensors