Now you can do full Stable Diffusion XL (SDXL) DreamBooth training on Kaggle for free under 2 hours.
Notebook and instructions : https://www.patreon.com/posts/88397937
Kaggle provides 2x T4 GPU.
With Kohya script we are literally doubling effective number of steps.
So when we do 80 repeating we do 160 epochs.
At the above image you see I did 160 repeating training.
So it was actually 320 epochs therefore overtrained.
Updated my Kaggle notebook post with accurate instructions.