Ask HN: Fine-tuning video generation models with reference video data

Hi HN crowd!

I'm interested to know if it's possible to create AI-generated videos based on other videos. For instance, let's say I have 10'000 reference clips, and I'd like to create some new clips that are in a similar style or convey a similar message.

I have pretty big knowledge gaps here, but I'm aware of the concept of "fine-tuning a model", so would it conceptually make sense/is it possible to achieve this via fine-tuning the model with the reference clips? Or would it make more sense to run the reference clips through some "summarization model" that creates a textual description of the video, and then use the collection of textual descriptions as input to the generating model?

Finally: is there a video generation model that can run locally on a beefy machine (e.g. Mac Studio) for some experimentation?

Thanks in advance.