this is an embedding that was trained on around 90 frames from marie antoinette taken from around 200 frames just randomly throughout the movie. I created flipped copies so somewhere around 180 total images, 424x424 because my computer can't handle 512x512 anyways what it seems to do is mostly dress people up like from the movie and maybe make the colours more pastel idk. it has the architectural style and the clothes pretty much down. I prioritized frames with kirsten dunst in them, so the women in your photos are going to be pretty blonde and stuff like that. idk I think it's pretty good. trained for around 700 iterations. will improve by culling the training material and training a bit more but for now seems good

marie antoinette sofia coppola

trained for like 700 steps on 180ish images from the movie

Files with SHA256: e3cb260815d45045a20bf5361ee307246cd3016b2776d994774985c74ea98f28