This is entirely an early Alpha version for now; it was made using a dataset of 7 pictures. If/when I get good high quality screencaps/fanart, I'll retrain and adjust the current images as well.
A couple of notes:
Avoid using
full bodyin the prompt, as it will heavily influence the backgroundfrom behindworks, but the pose can be a little rigid, as it is based on a single image.The details at the back of her dress are trained however, thanks to that 'from behind' image.
Naturally, dress and tiara and hair ornament details are not as strong as they could be given the small dataset, but the results are already promising.
Description
FAQ
Comments (5)
7 Images? Works great lol
Right? Granted, 7 carefully cropped and sized and modified pictures, but still, this is surprisingly results.
@Alienor_ What size do you train at? I'm curious about getting into training myself and any method that'll do it with seven images has my interest!
@Anistasiata Honestly, that was a normal 32/32. As it is I find 200mb to be a lot for characters, but in this case I didn't want to risk smaller.
I did make it single image batches, and high repeats in this case, obviously, but even then I overdid it - the backgrounds are very rigid/locked in and take a lot of prompting to remove... And that's after I cleaned up the images to try and avoid such problems.
Uh i probably can help you out with the screencaps, i've finished DT and i now have access to "FIGHT CLUB" toys XD
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.