I didn't find a concept LoRA for this to my satisfaction (eldritch anatomy and such), so I had a crack of making my own. After a few iterations and some trial and error, I've found something that seems to work most of the time.

Notes on use:

• Generated for Illustrious, not sure if it'll work on anything else but feel free to give it a crack

• Seems to work best at about .7-.9 strength. Higher and the anatomy can get a bit nightmarish. I've had some good results at as low as .6 and as high as 1.2, but you'll need to run a few batches to get some usable images.

• The LoRA can produce twins, but it's difficult attempting to get it done consistently. I've found turning up the model strength >1.0 can produce it if you're lucky. While I did attempt to tag it in my dataset, "twins" clearly has a pretty strong association on danbooru already. I also tried tagging the dataset with "two fetuses" where appropriate but it doesn't seem to work either. The LoRA seems to struggle with the anatomy too, so maybe in a V2 or something.

• If you're attempting to do this through clothing, you'll need to specifically note in your prompt that the clothes on the belly are transparent, otherwise it'll do this weird curtain thing with their clothing. Sometimes it'll work with lower strength or after a bit of fiddling.

• Seems to play nice with every other LoRA (Characters, concepts, didn't test it with styles). I tested it with, but no promises on anything specific. Probably won't work nice with anything that also does x-ray though.

• Works the best froim the side, mostly works from the front but definitely more anatomy issues, have got it to generate from the back decently once. Seems like the LoRA has trouble with orientation.

• Often won't take up the whole stomach, unfortuantely you just gotta play around with it until it works.

• You can kind of do middle stage pregnancy, try leaving out "pregnant" (since the checkpoint defaults to like 9 months) and just use "early pregnancy" or "small belly" instead. Careful though, it does weird stuff like place the x-ray where the spine should be, and the LoRA will struggle with anatomy as it gets smaller. You can get acceptable results (see above), but that took me somewhere around 30 tries.

• No support for furry stuff, alien eggs or other stuff. Doesn't really interest me so you'll have to make your own (unless you pay me hahahaha).

• No support for changing the skin colour as of yet.

• Mostly generated at 1536*1536, cfg 6.5, 28 steps.

Trigger word is "pregnancy x-ray".

Seems like I picked something kind of difficult for my first LoRA, so if you've got any tips or feedback it'd be great to hear. I'll probably make a version two at some point with a revamped data set with better examples and better captioning.