Just for fun — testing only.

I’ve uploaded two LoRA versions: one trained at step 2250 and another at step 3000.

In some scenarios, one performs better than the other. It needs more testing, but I don’t have much time right now.