RealSR VS DAIN

If you want to upscale live-action stuff, you need to load up some different models. For example RealSR or SRMD.

This tool allows for multiple drivers, you can look at RealSR or SRMD NCNN

Here is the last one I played with, but click the link above as there are newer models: https://github.com/baowenbo/DAIN

Not OP and unsure it would work well in this case, but I usually reach for DAIN to do a few frames of interpolation - https://github.com/baowenbo/DAIN

You may want to look in to DAIN https://sites.google.com/view/wenbobao/dain

Not sure if my eyes are playing tricks on me or if the video is slightly choppy. Maybe trying running it through DAIN to make it buttery smooth: https://github.com/baowenbo/DAIN

Nah this one is actually machine learning. There's plenty of machine learning interpolation networks out there. DAIN is just one of them. https://github.com/baowenbo/DAIN

This is DAIN https://github.com/baowenbo/DAIN

You could try an AI like DAIN but I shudder to think how much VRAM you'd need to handle 12k footage, you'd probably need something like an RTX A6000, and even then I doubt that would be enough...

If you've got suitable hardware, you could run the clips through DAIN or a similar AI interpolation suite to double the framerate - that'll get better results than Optical Flow.

This exists, it's called DAIN. There are apps you can download which processes GIFs into 30fps, 60fps, etc. https://github.com/baowenbo/DAIN