Perhaps there's an interesting ML research project there. Get a voice actor to record a bunch of lines, with facial capture, and then see if the AI can produce the correct facial expressions for a line it hasn't heard before.
Then, take it a step further by trying to produce the voiced line from a sufficiently detailed narration and transcript. Obviously GPT-3 is close to being able to produce narrated stories, so that would give you a pipeline that turns a vague story idea into animated speaking faces.
Generating the movements of bodies and props might be harder, but I'm guessing that in some CGI movies for kids they don't have any human characters so they don't need to capture any actor's body movements or have any physical props or a stage at all.
Then, take it a step further by trying to produce the voiced line from a sufficiently detailed narration and transcript. Obviously GPT-3 is close to being able to produce narrated stories, so that would give you a pipeline that turns a vague story idea into animated speaking faces.
Generating the movements of bodies and props might be harder, but I'm guessing that in some CGI movies for kids they don't have any human characters so they don't need to capture any actor's body movements or have any physical props or a stage at all.