I’m looking for adivse on what model architecture would suite the following problem the best.

I have a few complex animaitons that are each described by 70 variables changing over time. Their length can vary from one to many minutes. I can sample them for training data at any fps between 1 and 60 if needed.
I want to try to train a model on some of them, in order to see if I can generate similar ones based on a shortish input animation sequence.

I’ve looked at models used for mutivariate time-series prediction like autoformer and informer on huggingface. But my training data is not yearly/monthly/seasonal so im not sure if these complex models are most suitable. Their output is also a distribution if I understand it correctly, which I probably don’t need. And if possible I would like start with a simpler one.

Any advise is most welcome 🙂