Regarding questions about Unet,

#21
by LemonBranny - opened

Thank you very much for providing such a perfect model from your team, but I have some questions I would like to ask.
How many times can an image be learned before Unet starts to have problems (The maximum value of repeat divided by the total batch size)? What is the principle behind using a second training session to improve Unet in your project? I often encounter issues with Unet during training, such as blurry details, color blocks, and incorrect limbs. Is the second training session in your training method aimed at improving these issues? Can they be improved? What is the working principle? Is the second training session using the Finetune method? What is the difference between the Finetune method and Dreambooth?

Linaqruf changed discussion status to closed

Sign up or log in to comment