Can't have a conversation about inbetween frames without sharing this[1] Noodle video. Now granted the interpolation there is pretty rigid, and the goal of this dataset will be to train an LLM replicate hand tweening, but I can't help but feel there's going to be a massive uncanny valley.