My sincerest apologies for the nightmare fuel below. These are the progress pics for PornoGAN, an attempt at a generalized porn GAN. I've put together a dataset of (currently) 852,000 images to train on, but I don't think my 3090 is up to the task.

As you can kind of see, progress came to a halt around the 20th checkpoint, and besides *maybe* getting a bit better at making faces, it didn't really show any signs of improving, so I killed it.

With more GPUs and a lot of time, there may be better results to come, but I decided I probably have better luck training on specific scenarios. Separating the dataset out by position/content will probably give the feature extractor something to latch on to. Stay tuned for the results of that.