For me, it's fidelity of human facial animation that has a long way to come. There are teams doing a great job of it [1], but the labor/skill required to bring one high hyperreal facial rig to a game or movie seems insane. I think companies pursuing AR/VR applications will lead here.

ML creative tools stand to automate a lot of this imo.

[1] random example: https://snapperstech.com/