Cute character, I personally havent used it, but I notice a lot of LoRAs for OC chars trained with data made with AI being made lately. And I certainly get a lot of requests to create this sort of thing for people.

Maybe it would be best to create an embedding that you used to describe the char in your prompt for creating the dataset.

I think if you can produce the chracter with the prompt in any model, an embedding might yeild better or similar resuts. As an embedding is like a specialized group of tags condensed into a single tag. It also has a different affect than just normal tags. The prompt adherence feels incredible strong with it.

But I can imagine if you create a dataset for a complex OC that uses muliple loras to achieve its look. then creating a lora would be necessary.

So, maybe its worth giving a try.