This is my first attempt at training a model on SDXL. It was trained on about 600 ghibli style images for 20 epochs. When testing, I found it produces quite nice abstract style images and excels at anime type images with a solid prompt.
We'll be training more models over the coming days and weeks as we get to grips with what works and what doesn't. These models will be available over at Supermachine.art where we offer the best that we create and open source with 50+ models available.
Anyone interested in the more spicy stuff can check out Nudes.AI for which we have added a bunch of textual inversion characters previously. I really like some of the outputs from this model compared to the base SDXL and it shows a ton of potential on a small dataset and limited epochs so that's super exciting!
PS. I am sorry this is a CKPT file and not Safetensors. Seems to be a bug with OneTrainer currently where the .safetensors output is causing an error, ckpt doesn't. If anyone can convert / know how to convert to safetensors, please get in touch and happy to add.
Thanks and enjoy!