Spaces:
Runtime error
Apply for community grant: Personal project
Hi Team, I have created this Space for fine-tuning the Stable Diffusion model (supports the latest version SD 2.1) using a novel approach called LORA by Simo Ryu (He can be found over here - https://twitter.com/cloneofsimo). This space is based on the GitHub repo here - https://github.com/cloneofsimo/lora . This Space was a result of inspiration from one of AK's tweet and suggestions - https://twitter.com/_akhaliq/status/1601120767009513472. The main idea in the LORA approach of fine-tuning SD is the small size of trained model weights (only 3-4 MBs) in comparison to large sizes of Dreambooth fine-tuned and Textul-inversion models.
Awesome assigning gpu
Thank you for making this awesome space! thanks for the credit too!
1.it does not seem to be picking up any fine tuning. I tried 3 very different style sets, either all cropped to 512x512 or 734x734. some sets had no padding margins. regardles of alpha value there is nothing from the source images. also tried diff number of training samples from 4 to 20 with same results without fine tuning.
2. regardless how many steps i set it will always end up with weights points ranging from 500 to 2000 even if I set steps to 1000 or 3000.