How to train multiple subjects in Dreambooth

[jujpenabe]

I used to train models with multiple subjects by simply renaming the instance images differently (YYY (1).png, YYY (2).png...etc and ZZZ (1).png, ZZZ (1).png...etc) Is this still the method for training multiple subjects? Can I leave the captions empty or is it used to differentiate the subjects?

[nawnie]

still works keep text encoder at a low lr