I've decided to experiment and make a LoRA for both Polly and Kim from OMORI. The reason why I've decided to do this instead of one LoRA for each character is because if you try to combine them, then it will look more like a mess. Obviously I first tried to train Kim lora, and after succeeding I made lora for Polly, and then tried to combine the dataset and train lora on rank 16 and 10 epochs... aaand it failed. I'm not sure whether the issue was in small amount of training steps or under-parameterization (low lora rank), so I've decided to double both! And this is what we got, a working lora. This LoRA is compatible with the OMORI style lora I've made earlier for Anima!

Files with SHA256: 089618ca14dcce5259aa8dff5531ee47e111f4b5b9686e0a0c1bba7fd2ac2e4e