Jay Garrick (The Golden Age Flash) - v2.0 Showcase

A bigger dataset with more head closeups means this version is more likely to generate a helmet with the characteristic wings--but also left this model a bit more unstable than version 1. (Plus, the loss rate spiked way up for four epochs during training, during which the model went completely haywire.) I'm currently working on a version 3 that should hopefully do better. I've got some ideas for upcoming LORAs, but would love to hear suggestions or feedback!