My first LoRA training attempt. Using a small dataset, using autogenerated AI descriptions, using default tensor.art settings with reduced repeats/epochs. Even at low steps (~500) the training took 7 hours (yikes). The sample 2-second videos were abysmal, but the proof is in the actual pudding as they say, so I'm gonna burn some credits testing it.

I wouldn't use as-is until fully tested, as MMH3 costs are too darn high!