Back to the problem of multiple figures: I find this is fairly easy to arrange, so long as there are only two. I started with a hentai model (1) took out the irrelevant background and constructed the scene in Img2Img with the Lineart pose control and Realistic Comic Book - 1.0. This is a very simple and flexible model which can be used with most of the loras available. Unfortunately, the results often look like shit, as they did in this case. But no worries, I just changed the model to my old reliable, Sym-Pony world 2.5D, which produced a really stunning result (2). It’s like the AI could read my mind. But what if you want photorealism? For this, I turned to a high-end porn model, Mklan NextGen XL-NSFW (3) which after a couple of tries turned in something that could have come from a photographer’s studio (4).