CivArchive
    ReVulva - Q21 - v1.1
    NSFW
    Preview 149547619
    Preview 149547196
    Preview 149547865
    Preview 149547957
    Preview 149547410
    Preview 149547490

    ReVulva is a LoRA for Qwen Image 2.1 that’s meant to supplement the model’s deeper understanding of the female human anatomy as a concept, allowing for verbose and complex prompting while leaving its underlying comprehension of style and detail largely untouched. It was trained on a highly varied dataset of different subjects with completely different features, in an effort to show the model what a vulva can look like, so it doesn’t just copy/paste a single examples. Check out the attached images for a look at how to prompt with this LoRA.


    —TLDR—

    Strength: 0.8-1.2 (leave at 1.0 in most cases; ≤ 0.8 if combining with other LoRAs)
    Samples: 30-40
    CFG Scale: 3.0-5.0
    Sampler: Your Choice(any that work with QI 2.1)
    Scheduler: Your Choice(any that work with QI 2.1)


    —Changes in v1.1—

    • Better understanding of female genitalia with more detail and more natural blending with subject's surrounding anatomy

    • Manually-tuned block weights to minimize extraneous effects on color and detail fidelity—this is vastly improved over 1.04

    • Responds better to complex prompting (or no prompting), in most cases

    • Allowed the model to train for a stupidly long time, then rolled back to the last stable checkpoint for tweaking

    • Does not eliminate previous learning from 1.04; it just deepens and expands on that learning


    —Keywords Used in Training—

    vulva, pussy, labia majora, labia minora, clitoral hood, clitoris, mons pubis, vaginal opening, anus, asshole, anal indentation, perineum, pubic hair, and many more...



    —Tips for Optimal Use—

    • Responds well to complex prompting— be as descriptive as you like. Qwen’s VLM understands a lot more than just the trained keywords.

    • Be specific (or not)— if you just prompt a picture of a woman’s vulva or a shot between a woman’s legs, the LoRA will still trigger, but you’ll get Qwen’s interpretation of “neutral” anatomical features based on my training (i.e. typically recessed or indistinct parts). This can actually be useful in some cases where the anatomy is not the primary focus.

    • Reseed frequently— since the dataset is varied, re-rolling the seed can give you different feature combinations.

    • more samples + higher resolution = more detail | higher CFG + specific language = greater adherence to your prompt with less creative interpretation— play around with this until you strike a balance that works for you. I’ve discovered that the right combination is highly subjective to the scene/subject that you’re trying to generate.

    • Try to prompt your details early, rather than later. A close shot of the vulva of a 25-year-old female tends to work better than A close shot of a 25-year-old female... her vulva visible. Focus the model's attention where you need the most detail early on, and you'll likely encounter fewer issues. Again, the LoRA will still work even if you don't, but you might come across some problems with detail or consistency if you don't front-load your prompting as described.



    —Known Issues—

    • Occasionally produces unwanted results, but in most cases this can be fixed through modifying your positive/negative prompts, altering your scheduler parameters, or re-rolling the seed

    • The weights favor certain aesthetics over others. I made my training data as well-rounded and detailed as I could, but Qwen came away with its own preferences. Very specific prompting can sometimes help push it outside the box.

    • Very occasionally, parts may be missing (e.g. clitoris or anus is missing). If this happens, and you are really attached to a particular seed, just mention clitoris or clitoral hood in your prompt to remind the model to include it, or positively prompt accurate vulvar anatomy or something similar. Alternately, you can negative prompt inaccurate vulvar anatomy, though the effect is usually a bit weaker.

    • Can simplify anatomy in long and distant shots. Qwen does this with everything anyway, to some degree, but it can be frustrating if you're going for total realism. See my note on schedulers below.

    • May infrequently produce an upside down vulva when subject is in an inverted position (e.g. bending over, lying prone). You can correct this by removing the word from your prompt altogether (e.g. 'a shot from behind of a woman's vulva' turns into 'a shot from behind of a woman' or 'a shot from behind of a woman's ass/buttocks'—the LoRA weights should still come through) or by re-rolling the seed.

    —Image Workflows—

    I would highly recommend taking manual control of your schedulers and sigma values if you want to squeeze the most of out this LoRA. If you're using ComfyUI for generation, I'd also recommend enabling TAESD previews in your settings, so you can visualize where those details are filling in across your timesteps and tweak your scheduler accordingly.

    None of this is strictly necessary, and the LoRA works just fine with the standard KSampler node, but you can get a lot more from it if you prefer a hands-on, micromanagement approach. Mostly, scheduler management can be useful if setting up distant shots, or images with a lot of competing details. Sometimes, just raising/lowering your shift value can produce drastically better results.



    —Training Methodology/Rationale—

    This LoRA was trained using Ostris AI-Toolkit, using the native BF16 version of Qwen Image 2.1 (though it performs similarly using the INT8 quant for generation). My approach was to let the model learn at a moderate pace while only allowing it to make changes to its weights in small increments. The result was a very stupidly long, gradual training run that I hope helped deliver a deeper understanding of the concepts I was trying to impart.

    The end result was more or less satisfactory, and it's probably the best that I'm going to obtain with my current dataset. I’m training on consumer hardware, so realistically, without a blank check to a data center API that can handle a much-expanded dataset, higher-resolution images, and longer training runs, I’ve pretty much capped out in any practical sense with my current setup.

    My dataset images and captions were limited to anatomy only. No information about sexual positions, sex acts, or sexual language (aside from light slang) was included—not that I have a problem with any of those things, but that information just wasn't the focus. No individual likenesses were included in the training data.

    I'm likely finished with v1 and my present dataset, as I think I've gotten the most from it that I can. Any future updates will be focused on a v2 version with an expanded dataset (or maybe I'll take a shot at an all-in-one), but that's likely down the road a while.



    As always, enjoy and please generate responsibly. If you've read this far, thanks for downloading!

    Description

    • Better understanding of female genitalia with more detail and more natural blending with subject's existing anatomy

    • Manually-tuned block weights to minimize extraneous effects on color and detail fidelity

    • Responds better to complex prompting (or no prompting), in most cases

    FAQ

    LORA
    Qwen 2.1

    Details

    Downloads
    222
    Platform
    CivitAI
    Platform Status
    Available
    Created
    11/19/2026
    Updated
    11/20/2026
    Deleted
    -

    Files

    ReVulva_v1_1_R32_Q21.safetensors

    Mirrors

    ReVulva_v1_1_R32_Q21.safetensors

    Mirrors