Create Audio Instructions Based on Video?

Can Gen-AI create voice instructions on how to do something?

It’s easier to record a video of a task: change a flat tire, juggle, demo yoga — without also speaking.

Mediapipe can detect the movement someone is carrying out. Can another model use this inference to generate audio instructions on movement?

Link to videos with movement detection:

Audition 6

Original video: https://www.youtube.com/shorts/DnDH4OshzB4

Another example with a juggling video:

Juggling Movement Detection

Original video: https://www.youtube.com/shorts/czM2Tib4QAU

Code: https://github.com/Anudha/Yoga/blob/master/MediaPipe_ForYogaPose_GenAI_Public.ipynb