Extreme zoom-in videos are something that have gotten a little publicity with some videos made by Jesse Martin.

Now Google and University of Washington have created a text-to-image model for extreme semantic zooms for consistent multi-scale content creation

Check out the Research paper here.

It seems like this kind of technique that over-arches the generation from one topic to the next might be very useful in maintaining continuity relating to temporal stability.

Articles:

• https://www.marktechpost.com/2023/12/08/researchers-from-the-university-of-washington-and-google-unveil-a-breakthrough-in-image-scaling-a-groundbreaking-text-to-image-model-for-extreme-semantic-zooms-and-consistent-multi-scale-content-crea/?amp