retrain the model with own images using dreambooth
Thank you for your work,
but I can't retrain the model with own images using dreambooth
Has anyone managed to do it?
I haven't tried it using dreambooth, but I was successful using a hypernetwork to train a custom set of images. Here is a good video explainer: https://www.youtube.com/watch?v=1mEggRgRgfg
(fwiw, I intend try dreambooth, too, but I don't have a PC that can handle it, which makes the install more time consuming. I will try it when I have some free time, and if it works I will post an update)
JTC
I did it, once, but I could never make it work after that
It works, and I advice you to leave the token field empty, only give Huggingface token it if prompted. I had so many problems that I didn't realize were related to that.. At least I think.
It works, and I advice you to leave the token field empty, only give Huggingface token it if prompted. I had so many problems that I didn't realize were related to that.. At least I think.
thanks friend, what retraining settings are you using 2000 steps? and what number of photos? Thank you, if you want we can talk more fluently on telegram
The number of steps are directly proportional with the number of new original images.
This guide suggests 100 steps pr/image:
https://bytexd.com/how-to-use-dreambooth-to-fine-tune-stable-diffusion-colab/
I suggest using more than 50 images, but it depends on what you're doing. If you're training with a bunch of images of yourself, for instance, make sure to do many different poses.
Google collab is great for training. My GTX1080 (8GB Vram) sure isn't enough for training, although it's decent enough for generation.
Keeps in mind I just followed the guide and experimented a bit, and pulled out my hair for a few days. So my guess may be as good as yours.