search

kohya-ss / sd-scripts

Apache License 2.0
5.27k stars 876 forks source link

[Question] What is the different between dreambooth and fine-tune in sd-scripts #568

Closed xarthurx closed 11 months ago

xarthurx commented 1 year ago

There is only Japanese version of fine-tuning, and usually people refer dreambooth to "fine tune" native stable diffusion/other diffusers.

So what is the main difference between the "dreambooth training" and "fine-tune training" in this repo? When I make a decision, what are the criteria I should refer to?

From my rough understanding, for dreambooth, you can train some image for multiple times (with folder name modification). And for fine-tune, is this for tuning the model with many images in the similar class?

Really confused about this.

kohya-ss commented 1 year ago

Sorry for the documentation. The main difference between the two is that the DreamBooth method can use regularization images and class tokens.

There is documentation translated by darkstorm2150. This is chatgpt4 translated, but seems to be of help. https://github.com/darkstorm2150/sd-scripts#links-to-usage-documentation

goodlux commented 1 year ago

Thanks, this is really helpful