Decided to take a stab at Stable Diffusion...here's what about a week of practice has gotten me.

                                                       

For a week, she’s pretty solid

Yeah after doing about 100 iterations within about an hour or two it started figuring out what prompts meant and kept getting better. I wish I could get more detail out of it and better skin resolution. Also if she’s got her belly exposed it takes “popped belly button” as a literal “button” like on clothes or on a machine. not a human navel. Unless I need to specify it as a human navel. Still figuring that part out. Hoping to get things more detailed as it progresses.
I’ve done a lot of work in DAZ3D over the years and thought I might jump on this band wagon. Hoping it inspires some creativity on the 3D side of things.

If you're using base sd 1.5 you have the option of looking for more models that are more suited to photorealism in their outputs and try your hand with those. Unfortunately there's not a lot of content out there that's focused on both photorealism and also bellies, most of that is being used only in the anime models. There's also merging models, trying to find a level of photorealism that you're happy with, or at least a model that's receptive to it while also trying to extract what you want from one of the models that's better suited to bellies that are out there online. Also LoRAs are very popular, it's probably very possible to train one on pregnant bellies if you're willing to take a dive into training.

I’m using easy diffusion, which is based off stable diffusion 2.5 I believe or 2.something… can you recommend any LoRAs that might work? I saw one user on here who was getting like super detailed bellies and I was like wow. I’m also looking for fingers to be a little better than they are. But like I said I’m wondering if what I do here can translate to 3D in anyway. Maybe like a compare/contrast thing… I don’t know

Unfortunately all of the LoRAs that I know of are trained on 1.5 and won't work on 2.X, so you'll have to do the heavy lifting with prompt work yourself, or switch over to 1.5 models. For 3D, it depends on what you mean. I know work in Daz3D can be translated very well into SD thanks to controlnet and img2img.

Here's some more playing around...
                                                                                                                   

iirc Automatic1111 extensions exist to generate depthmaps and meshes from images - and it doesn't have to be an AI image, you can use any image as an input. I've only played with the depthmaps so far but I can see the potential there to work back and forth between 3d modelling and AI generators and should probably at least see what can be done with Daz3D, here's hoping it's a less steep leaning curve than Blender.

I felt like I'd hit a brick wall for detail and then better photorealism models came out. They really do make the biggest difference but it's also a delicate balancing act of the other parameters, not just the prompts. I merge my own photorealism model but almost anything that represents itself as photorealistic will deliver good quality results with the right prompts and settings. Using img2img also helps a lot for experimenting to get the best quality.

Easy Diffusion can handle both SD1.x and SD2.x models, so you can use the appropriate Loras with the appropriate models. All of the pregnancy-themed Loras/models on civitai are either for hyperpregnancy and/or Anime style. I haven't really bothered with the Loras, they work with photorealism using reduced weighting but it's a different kind of not quite right for me. LadyBabyloin also has a link to their own pregnant teen-tuned model in their AI pregnant teens thread,  while it seemed to boost realism when it was shared I'm unable to integrate it to improve pregnancy realism with the latest photorealism models.

I've been doing this since the first month it was available and still haven't cracked the bellybutton puzzle yet either (though I only noticed the direct connection to actual buttons recently), Training is clearly the best solution going by bo27's pics, but I'm too caught up in trying things that don't take as long.

Hey all, still working on some stuff. I think I’m starting to get the training thing down. One thing that seems to be hit or miss is a belly that looks normal. Still trying to master the belly button thing, can’t seem to find the right prompt to achieve that. I am also running into the “butt belly” I call it where there is a line or butt crack where a belly should be. Anyone else run into this?

oh yes a lot, or the navel literally becomes an anus. Or the bellybuttons are positioned as though they were regular buttons.

I've had a lot of success recently using alternating words in the prompts for belly identifiers e.g. ([Pregnant|preggo|pregnancy|maternity|babybump]:0.9)and also using the anime pregnancy Loras at low weightings e.g. <lora:pregnant:0.4>, It's still not as realistic as a trained model, and with the Loras the navel can end up looking painted on if the weighting is too strong,  but the positioning and appearance  have been vastly improved imo compared to my earlier pics. Most of them prior to this week have had manual editing done to the navels; the latest ones could do with a touchup but they're so much closer to realism that I'm not going to bother unless I'm working on a specific concept.

I still get a lot of results with a small split at the base of the belly especially when using keywords for bigger girls, sometimes it can be edited, sometimes it's too hard to fix.