UPDATE FOR THE V2 :
The generations are overall cleaner, there is more variations understood by the locon when it comes to belly shapes and "firmness", let's say. It's more responsive to the prompts related to the belly. Some expressions that were corrupted previously aren't anymore, also. Still not perfect obviously, notably because the style of the locon still often overrides that of the model chosen.

This is yet another locon aimed at creating vore bellies. This one has particularly been trained on being able to render a variety of different poses that might be suitable for the vore fetish.

This is a v1, so it's highly flawed although it has potential. The style of the models you choose will be heavily impacted by the lora (for which I suggest strength 0.7 or 0.8), and there are often artifacts that make it so that several generations are often required to find some clean ones.

That being said, a vast amount of poses and angles are possible. The artstyle seem to vary depending on certain prompts you use, so sometimes you might want to adjust it all so you can settle on something clean.

View the examples posted on this page to get an idea of the prompting required. I used a lot of specific captions in my dataset in order to influence the generations. The prompts related to the kind of "belly bulge", however, do not seem to be responsive, or at least not the way it was aimed at. Those prompts being "belly bumps", "detailed belly bumps" and "soft belly).

"Vore belly" is the trigger word. I often need to put it at 1.3 strength to really create interesting bellies. For the upscaler, I use R-ESRGAN 4x+ at 0.2 denoising strength.

I'd like to achieve a locon that does not disturb the original artstyle of the model used. That seems to be above me, haha. It was the same thing for the last locon I created from pictures of a large variety of artstyle, and the aesthetic that resulted was mostly identical to that one here.