Trained triggers:

1. cellulite: creates deep skin dimples and a wrinkled skin effect, often seen on mature woman.

2. thick thighs: creates wide thighs below the butt cheeks.

3. wide hips: makes the butt cheeks wider, and the hip bones compared to the waist.

4. blemishes: adds a form of pimples, which can look like a rash sometimes. Use with caution or not at all.

5. Chubby: Makes the woman plumper and creates a mature body type.

6. Obese: makes the woman overweight and very curvy.

7. Shiny oiled skin: makes the skin glossy and oily, with specularity.

Testing:

I have tested the lora to a peak strength of +1.5, and haven't had the need to go above that to reach the desired results, and haven't had any distortion or deformations yet. So it can be used as a slider. Feel free to experiment.

the lora works best when the triggers are used in a array, seperated by comma's: e.g: cellulite, thick thighs, wide hips, blemishes. However, you can still use natural language for the rest of the prompt. My prefered method is to start with the string of triggers, followed by the rest of the prompt.

When using in the mix with other lora's, increase the strength.

If you want to add tattoo's, or any form of details, add cellulite to the natural language prompt, E.g: she has a tattoo of "xxx" on her cellulite.

It can do bodyshapes other than obese or chubby, you can use for example: hourglass bodyshape figure, or: fit muscular with only a obese lowerbody, thin waist legs knees arms. So quite alot is possible if you phrase it right.

With a very long context length you can either move the triggers to the beginning of the prompt, and if that is not sufficient, increase the lora strength after that.

When using cellulite as token, skindentation is automatically applied as well, so no need to prompt for this.

Notes:
I did not train on tokens such as: mature woman, or elderly, etc.etc. so it's highly versatile with all ages.
I used comfyui, my prompts and basic workflow is attached to the images, just drag'n'drop the images in comfyui.
The images in the dataset exist out of mainly nudes, lingerie, and thight mini outfits, in amateur settings. It might be able to do other things too, but I haven't tested this extensively yet. So don't expect to see cellulite through jeans or a long dress.
Don't post any images that violate the law, follow the TOS, and keep the platform pleasant for everybody. I used real photos from pre 2020, which I had saved to a harddrive, so no AI images are in the dataset.
And have not yet tested with 2D, 3D, or other media types.
Use words such as "photography" for realism, not "realistic" which implies semi-reality.

For the examples I used Krea2 Basemodel: realismByStableYogi_v20TurboINT8, CLIP: qwen-3-vl-4b-heretic_int8, VAE: wan_2.1_vae, inference steps: 8, sampler: euler_ancestral, scheduler: sgm_uniform, resolution: 1080x1920, 1080x1080.

Hit the like button on the model page, and on my images if you enjoy what you see! I'm all out of buzz, and every bit helps so I can train more models! And if you download the lora, posting a image to the page grants me 50buzz per user! So I would appreciate it alot! All donations are welcome too, but don't feel obliged if you cannot afford it like me! xD

You can sell your images, and do whatever, but if you make merges and big profit, sending a tip would be a cool gesture. ^^