It ended up being a style LoRA that basically just added glossy skin to Riz's art style. Even though the training images consist almost entirely of characters with sharp chins, the resulting LoRA produces round faces. I wonder if it is being influenced by the small number of round-faced character images included in the training data, or if the images of side views and the backs of heads, which make up more than half of the training data, fail to properly capture the character's facial contours... Making LoRAs is tough.