Possible to train once and then add totally different person later to the same model?

[jysung]

Is it possible to

1. train first on 30 images of SubjectA for 3000 iterations and save the checkpoint file
2. load colab with that checkpoint file, and then train a new session with only 10 images of SubjectB images for another 1000 iterations?

[TheLastBen]

Yes you can, I recommend setting the text encoder to 30% and the steps to +3000, 4000 is a sweet spot