Lora of Yuri, a character from the YouTube web series 'Etra-chan Saw It!', for Anima Preview 3 and derived/fine-tuned models.
It's been a while, huh? I've been around since the Stable Diffusion 1.5 days, and now it turns out we've finally moved past the traditional Stable Diffusion architecture. Wow, that's crazy. Anyway, this is one of my first LoRAs trained on Anima—at least, the first functional one—and it's a character I had already made a LoRA for back in Illustrious. I’ll be taking care of porting all my previous LoRAs to this new Anima model—or at least the ones where I still have the dataset, since I’m way too lazy to make new ones from scratch.
I noticed some curious things while trying to find the perfect balance for this LoRA by experimenting with tags. At first, I used a prompt that worked for me before, which included the 'bangs pinned back' tag for the character’s bun hairstyle. But for some reason, it kept adding some hair falling over the character's forehead, which wasn't the plan since the original character doesn't have that detail. I tried fixing it in several ways—adding 'forehead', 'topknot'—but nothing worked until I realized the 'bangs pinned back' tag was the culprit. My reasoning is that this tag probably had many images in its dataset following a pattern where some hair still fell on the forehead instead of exposing it completely. I was losing my mind thinking low-quality images had slipped into my dataset and that I hadn't filtered them properly.
Anyway, just some trivia nobody cares about! I appreciate whoever is reading this taking the time to do so, haha. Feel free to do whatever you want with this LoRA.