Making AI characters that stay the same person across every angle, outfit, and video clip... Turns out that's harder than it sounds. LoRA training, endless prompt tweaking, way too many test batches. If it looks consistent, I probably broke it five times first.