I found I go between .25 - .50 for most things. Again I created several check points, so through out the process I might toggle back and forth between what model to use.

spent way too long photobashing again. v2 + sd1.5 can get something resembling a sort of kitbashed realism, but it's time-consuming. v2 really does amazingly well with shapes though

I started training a new version of my hyperpreg model on the latest Stable Diffusion 1.5.  Should be able to get even larger bellies than v1 of my model.  Will aim to make this one cater towards realistic images as well.  I'lll post a new thread when its done.

Super early version of the new model shows promise in belly size (which should be somewhat adjustable), but I need to retrain it to make it better quality visually, and make it less artificial looking.  This one was way overfit

(November 2, 2022, 11:12 pm)thojm I started training a new version of my hyperpreg model on the latest Stable Diffusion 1.5.  Should be able to get even larger bellies than v1 of my model.  Will aim to make this one cater towards realistic images as well.  I'lll post a new thread when its done.

Super early version of the new model shows promise in belly size (which should be somewhat adjustable), but I need to retrain it to make it better quality visually, and make it less artificial looking.  This one was way overfit

looks very promising

Almost forgot to mention I finished training a v1.1 of the hyperpreg model. It's very similar to v1 except it allows you to generate larger bellies.

The torrent can be found here among the large list of everyone's models
https://rentry.org/sdmodels#hyperpreg-v11-realism
keyword: hyperpreg (same as before)

See above for early results, the results are better now, I'm just too lazy to post. Hopefully I still managed to keep the realistic look and feel like the v1 model

Thanks for sharing your model!  I played around with it a little bit, but with only moderately satisfactory results.  For me, he keyword "hyperpreg" only seems to create regular sized bellies.  Am I missing something in the prompt?  It works decently with IMG2IMG but the people end up looking very like CG 3D renders.  But I like how it doesn't generate the annoying deformed hands I get a lot before I used your model.  Thanks again!                

I usually under train my models a bit to keep it from becoming too CGI/Digital illustration like, so you should prefix the keyword with "huge" or give the keyword more weight to emphasize it.  Sometimes "huge hyperpreg pregnant belly" works well too.


"a beautiful woman with a huge hyperpreg" raw results below

Okay thanks I'll give it a shot!

I'm starting to get some better results now!  The bellies seem to get smaller the more prompts I add. Thanks for the tips    

Yea, that's definitely the case. The more prompts you add the less attention the keyword gets. It helps a little to make the image a little wider than it is taller too. 576x512 or 640.