V2
Bellies seem to come out bigger, and the lora seems to have less unwanted impact on other elements than with V1, at least from my testing. As before, the main phrase is "swallowed someone whole" worked somewhere into the prompt... but there are a lot of phrases that can help push it further (see the showcase image prompts as good examples).
I'm excited to see what others are able to generate with this. Together, we can build the men we all dream of being gobbled up by. (everyone has that dream, right?)
Note: One thing I don't like (and I'm hoping to train against in future versions) is the way hair on the abdomen tends to be a defined line down the center. If you're seeing this a lot, you can fight back against it a bit by either describing the belly as smooth/hairless, or by explicitly describing the texture of the hair on the belly and how it should look instead.
Human Male Vore Belly
This is my first attempt at a LoRA, so hopefully I've done things correctly. This was trained on FLUX.2-klein-base-9B, but most of my generations have been with FLUX.2-klein-9B (the distilled version), and I seem to get fairly good results.
Triggering
Since FLUX.2 uses a text encoder that is pretty language-aware, I picked a keyword that is fairly natural. This LoRA is trained with prompts where the phrase "swallowed someone whole" is worked somewhere into the description. Additional descriptions of the belly as being massive, distended, etc, also go a long way.
Description
Comments (4)
Looks great! Any chance for Illustrious or Z-image?
Honestly, I'm not sure. I've been trying Illustrious for months but I just can't manage to train a lora to get the desired effects without the generation being cursed as hell. Now that I've got things figured out for this model, though... maybe I can use it to create better datasets for training the others? I'll add it to the list of stuff to play with.
I'd be remiss to not point to @oilio2 (in case you or anyone else reading this thread hasn't stumbled into them yet), who has quite a few models that work well in Illustrious (and really any Pony-derived model, if you're patient with tweaking weights and prompts). My prior realistic(-ish) generations were mostly a combination of those LoRAs and Cyberrealistic Pony.
@tevstv788 Thank you, I'll take a look.

