Replies: 3 comments 2 replies
-
Hi @daeunni. If you haven't already, I would like to recommend you to read this blog post: Training Stable Diffusion with Dreambooth using 🧨 Diffusers. They say "... but it's hard to select the right set of hyperparameters and it's easy to overfit." |
Beta Was this translation helpful? Give feedback.
-
Hi, @StandardAI, thanks for your recommendation. I will try to modify the learning rate. |
Beta Was this translation helpful? Give feedback.
-
Maybe @linoytsaban @apolinario has some suggestions? |
Beta Was this translation helpful? Give feedback.
-
Hi, I am struggling with the editability problem when running DreamBooth + LoRA.

I manually collected the plushy images from Google, and put them to DreamBooth + LoRA. (I utilized default setting)
However, I got pretty poor results like below. Most of the generations don't follow the prompt, and they seem to be overfitted by few-shot user images.
Does anyone know why it happened, and how can I address it?
Also, does any papers address these overfitting issues?
Beta Was this translation helpful? Give feedback.
All reactions