As a quick hack test, try this (set project to not use worker mode):
Add a usermedia element to your layout.
In scripting use:
// Get the video element (assuming only 1 video element present)
globalThis.myV = C3UserMedia_GetVideoElement(0)
globalThis.myV should point to the usermedia object, which you use for your tensorflow pose.
If it works, request to Ashley to add an official scripting SDK for Usermedia to get access to the usermedia element to import to other APIs. I recommended only using official scripting/SDK API calls.