This was a model I created by accident when trying to train on a similar concept (2 women standing in their underwear.) I trained it on 11 pictures of Asian and Caucasian 20-30-something women standing in their underwear in groups of 2 as well as another 11 images of extremely beautiful faces. For some reason, the women produced by this model seem to be having a ton of fun in a state of undress (though not completely naked unless you specifically prompt for that.) This model also seems to produce quite a few extremely beautiful mixed-race Eurasian women, having been trained on both Asians and Caucasians, even when you don't prompt for it.Depending on your aspect ratio, you'll usually get 2 women in the photo though I've seen 3 and sometimes more.I haven't tested it thoroughly and don't plan on further developing this one specifically but check back for a link to the main model which I'll post here when it's complete. Use the default prompt I suggested here (or at least mention some of the keywords in it) to get similar results.To get started, try the following settings:Prompt: 2 barefoot women standing in their pantiesNegative Prompt: [long legs], [skinny], [fat]Sampler: Euler AHeight: 1024Width: 512 (or 640)CFG: 7.5Steps: 20High Res: On, Denoising = 0.5(Note: after the recent update, use an initial resolution of 384x768, an upscaling factor of 1.33333, and an upscaler of Latent.)Sometimes, the faces turn out blurry. If that happens frequently, make sure you're using the SD1.5 VAE and have Restore Faces checked. If those still don't solve it completely, add perfect face (or sweet face, sexy face, hot face, etc.) to your prompt so that it tries to create a high quality face it was trained on (though it'll often be at the expense of higher quality elsewhere in the picture.)Since this was trained on <a target="_blank" rel="ugc" href="https://civarchive.com/models/1276/dreamlike-photoreal">Dreamlike-Real</a>, it supports <a target="_blank" rel="ugc" href="https://danbooru.donmai.us/tags">danbooru tags</a>.

Files with SHA256: 4f43162b5b86ca5d398c02f9bcf4159d1eee53b67be3d0676d3e9de91f1cd5bf