Qol improvements

[nawnie]

not sure if you looked at the edits i made for the guy comparing your dreambooth to another, i removed the ton of down load links though i may add them back for me torrent and url download, all this does is assings variables to the batch steps gradient steps and learning rate, when i get a moment ill also be adding a change to the warm up steps as i personally find it best to do 2x the amount of subjects to ensure each one gets at least 1 pass before going up to a higher rate.

learning rate is set with a drop down of 3 settings, your default 2e-6 also added 5e-6 for animation this is a great start. and a 5e-7 for fine tunning with large amount of steps, all this is explained via markdowns, i believe this allows for a greater range of fine tuning as well even if it is unneeded with smaller data set.