I spent about three weeks trying to get text-to-video to produce consistent characters across scenes for a short project. Different prompts, different models, different seed tricks. Every generation gave me a slightly different face, different body pro…