I'm back!

Hey everyone! This is a new attempt at training a character LoRA, this time for Hekapoo from Star vs. the Forces of Evil.

I followed most of the recommendations from my previous model (better dataset cleaning, standardized tags, proper character LoRA settings — dim 8, no caption shuffling, keep_tokens on the trigger word), Natural tag for a character, and honestly, the results came out better than expected.

It's not perfect yet, but I think it's getting there. Feel free to use it and if you get anything interesting, please share it here — I'd love to see how it performs across different prompts and styles.

Trigger word: hekapoo

Thanks for testing it out!