Novel training technique so that can train the Dreambooth consecutively without forgetting prior learned concepts

[kohweiteck]

I came across these novel training techniques to continuous diffusion:

https://jamessealesmith.github.io/continual-diffusion/

In short, based on interpretation of the paper can train the multiple concepts in continuously sequence without the model suffers what they had learn prior . Which currently Dreambooth suffers from currently.

.Just put it here, for sharing so that @d8ahazard may take note of this once it is available.