This time I made the model with only the trigger, i.e. no captions. Neither version really require a trigger word but they do refine it a bit... idk how to fix that atm. This edition could have probably benefited from one more epoch... I'll try that later. I think this version is more flexible and has way less quality/artifacting than the previous. Upping the strength to 1.2 make the face/body a bit softer but can also distort things a bit.
Also, all images were generated with ComfyUI rather than SeaArt and it's prompt and quality guidance.
