A model depicting the character Tyranny from the Critical Role Campaign 4.

There wasn’t a character LoRA for her, so I made one myself.

Is it good? Probably not.

Lora is also likely trained on too little dataset.

Version V1.3 is trained on a dataset roughly eight times larger. It should perform a bit better. Not all styles and checkpoints understand “horizontal pupils” correctly (change strength appropriately). The earrings are still inconsistent.