This is my testing model as I work out how to best train sdXL, so it was trained on a much smaller dataset than the 1.5 sd version for testing speed so that being said this model will be pretty limited with characters and pretty much anything that isn't a solo simple pose.
Description
Starting from the ground up on a new 70k dataset, results are looking promising as it seem to be learning characters and artist's names faster than I would expect so maybe training won't take nearly as long as I first thought.
Still early and in testing so expected some weird images from time to time.
also the "real e621" trigger word is no more as it seemed to break training.
FAQ
Comments (3)
Sorry for no updates in awhile, have many weird training bugs ie. pictures generated have very low contrast and clarity or are super over exposed.
will keep testing, but have had to start over :(
Is there an eta on an update?
@redstonehero well I have a model I could post thats about 40% better than the 1 posted here, but I trying to get it to at least 100% before posting an update. Just kind of my perfectionism mentality, it really eats me alive if I post something that I don't feel is fully up to my standards
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.