I only gather about 25 images a night, and only if I have enough free time. I train models while I sleep, and I generate new nudes in the background while I'm doing my "nightly browsing" each night, so overall it doesn't take too much out of my time. I had more than usual of Kate Beckinsale in the above post because I was also generating some while the Game Awards were on last night lol, so that added a lot more generation time.Wow, that’s some dedication! So at what point did your hobby become your job and your job become your hobby
In general, I find that stable diffusion models have a hard time generating more than one woman in the same picture, unless they are of a different race. If you try, you either get the same woman copied, or you get a blend between the two women. Man and woman though, you can do with some success, although I have seen some blending of features there too. Since combining two women in one image is difficult to do, I tend not to train different women into the same model for that reason. Just focus the model on being really good at doing one celebrity.
The best way I can include two different women is by taking two different trained models, merging them together, and using a prompt with both their names in it to generate an image with two women in it. Both of these women will appear to look like a merge between the two women, but one will look a little closer to one of them, and the other will look a little closer to the other. I then use inpainting to replace each woman with the one I think looks closest, using the original individually trained models. I tend not to do this very often because these take significantly more work, but here's one example I did with Kristen Bell and Mila Kunis together:
View attachment
