Other Flux models than D1

Currently the on-site training only supports flux dev, but it’s been shown to have substantially less utility than Flux1D2pro and the like. Letting us set custom flux models would be helpful with this.

Helping us by allowing us to continue training from other loras. This is a pretty normal feature with Kohya and I use it regularly, so continuing training from earlier Loras would definitely be a pretty useful trait.

I find the most useful utility comes from continued training on the same lora with flux. I’ve been shoving data into the same lora for over 2 weeks now and it’s only getting smarter.