Future plans:
I have a new neoteny dataset ready and waiting for z-image base to arrive. This is a much smaller project, but I'm excited about it all the same, because I now have even more beauty in it, a bigger set, and some examples across genres - like Victorian and medieval fantasy, but this time all photographic quality. It also has a few ethnicities for variety in the mix. This should be a go to for beauty on z-image when trained. I'm also some substantial way through my mega project, probably going on a year when done, fully photorealistic fictional settings and characters. The advent of z-image means I'll likely do this one in two phases - first a train on the data I have, for a 0.1 or 0.2 version of the mega LoRa. Then I'll use that to pass the smaller subset of higher quality synthetic pictures I have. I also still have about 50 pictures to improve clarity on, and label. All going well, I hope to make this not just a lora, but a full base model. Fingers crossed. It'll come out either way. It should be capable of things no other model is, because the dataset is photographic quality, but modified for extreme clarity and elements changed AND the dataset is extensively captioned, describing everything in scene (which is manually edited, including things only humans can pick out, like beauty) - essentially it doesn't exist on web, and can't be generated with crawling or automated captioning. When I do release it, I'll write an article about all the steps involved in making the dataset, should anyone ever want to take on something so promethean.
