NaturalBeauty Qwen Nudity is a photorealistic Qwen-Image-2512 LoRA for creating nude and topless images of beautiful women.
To trigger the LoRA, add
nakedortoplessto your prompt. (bottomlessalso works!)You can also add
breasts visible,nipples visible, and / orpussy visibleto accentuate things further.
NaturalBeauty enables Qwen to reliably generate naked and semi-naked photos with correct anatomy, and with a more photorealistic look than the default Qwen-Image-2512 base model.
This LoRA is the result of a two-year project to tag 35k+ images from a diverse dataset. No watermarks, no tattoos; just beautiful images of beautiful women.
Tagging
NaturalBeauty Qwen Nudity can do more than just nudity! In addition to naked and topless, try adding any of the following extra details for more control:
{a caucasian|a black|an asian|a latina|an indian} woman,
{standing, wide stance|standing, feet together|kneeling|squatting},
{naked|topless|bottomless|breasts visible|nipples visible},
{bare feet|high heels},
{short|medium length|long|very long} {braided|braids|crimped|curly|ponytail|straight|wavy}
{black|dark brown|medium brown|light brown|chestnut brown|auburn|orange|red|strawberry blond|light blond|medium blond|bleached blond|platinum blond} hair,
{neutral expression|slight smile|full smile},
{closed|slightly open|open} mouth,
{closed|open} eyes,
looking at camera,
{tiny|small|medium|large|huge|gigantic} {perfect|pendulous} {fake|natural} breasts,
{arms by side|hands behind head|hands in hair|hands cupping breasts|hands on thighs|hands on hips}For example:
a caucasian woman, standing, wide stance, naked, long wavy medium blond hair, slight smile, closed mouth, open eyes, looking at camera, huge perfect natural breasts, hands behind headNaturalBeauty Qwen Nudity is optimized to produce good results without needing a negative prompt.
Recommended generation settings
NaturalBeauty Qwen Nudity is trained to generate full-body portrait images with a size of 1056x1584 pixels. Other 2:3 aspect ratio sizes should also work.
All of the example images were generated in ComfyUI using:
…with the following settings:
Sampler: Heun
Steps: 4
CFG: 1.0
LoRA model strength: 1.0
Natural Beauty works great with the 8-steps Lightning Lora too, and also with just the base model by itself (switch to steps: 20 and CFG: 2.5 when using just the base model).
All of the example images for this model have a ComfyUI workflow and instructions embedded inside them. Just drag one of the sample PNG images into ComfyUI and generate!
Training settings
35,507 hand-tagged images (naked and clothed, no tattoos or watermarks), plus flips
Extensive tagging for ethnicity, breasts, nudity, clothing, leg pose, arm pose, hair, eyes, mouth, jewelry, piercing
1x H200
Constant learning rate of 5e-5 over 6 epochs at full 1056x1584 size with batch size 1
64-dim bf16 LoRA export from fine-tuned base model
Trained with musubi-trainer
Notes
This LoRA is a follow-on to a FLUX checkpoint I released back in November 2024, named NaturalBeauty FLUX Nudity.
Description
Initial version for Qwen-Image-2512.
FAQ
Comments (7)
The example images look like Z-Image base, very over saturated.
Z-Image Turbo is undersaturated. Base and Qwen are just normal in comparison. ZiT underwent RL finetuning to steer it towards that "filmic" look like FLUX.1-Krea. ZiB and Qwen are flexible base models so you have to prompt for that if you want it.
My monitor is calibrated and the saturation looks fine, they do slightly look overexposed though. Qwen 2512 is definitely more cinematic-looking by default but this was trained on 18,000 images and porn is not commonly graded in that muted style.
NaturalBeauty has a tendency to highlight / lighten the front of the person in the foreground, mostly because that's how these kinds of photos are typically lit. That might be why things look more over-exposed than the default. People certainly end up more front-illuminated than base Qwen, at least.
As for any color saturation: that might just be my preferred prompting style. I tend to prompt for lots of color and lighting style by default. I don't think more color saturation is introduced by the LoRA itself.
They are also only four steps, and it's probably based on the prompts as well. I don't even use/like qwen and these are pretty nice
... are you looking for some sort of apology?
Based on all the sample images, you've used about 2000 images of the exact same woman. Variety is better than "perfection."
Oh, interesting - I'm not seeing that myself from the samples. The training set for this LoRA was carefully curated to be as diverse as possible, to avoid that kind of repetition.



















