Can Object Capture convert video to 3D model?

Photogrammetry uses still images (and additional data like gravity/depth) to find similar corresponding regions on several images simultaneously and reconstruct 3D object from it. It cannot be done on video stream because you just have no next "image" so to speak... Video stream usually contains only few full keyframes and stream of "changed pixels" to build next frames from previous keyframe and changes. That's why you might see "stuck" potions of videos sometimes when connection is weak or file is damaged. Frozen potions of screen "overwritten" by next frames with nice "glitch" effect. Until next keyframe with full picture. That's why photogrammetry session cannot compare several frames and detect corresponding regions to reconstruct 3D point-cloud and build a mesh for it.

You might convert you video into image sequence (command-line utilities or any video editing software) and feed its output into photogrammetry session (i.e. command-line HelloPhotogrammetry example).

It would be nice to have iOS app capturing video along with ARKit scene so each frame would ALREADY know its position in space. It will significantly simplify reconstruction process (and affect quality). There IS such app for camera path (in AR) along with video (to use in CG/compositing). But to combine both approaches is a bit tricky. Definitely it would be done in future. It's obvious that we need a synergy between ARKit and RealityKit (object capture/photogrammetry). But for this very moment Photogrammetry is MacOS only API while ARKit is iOS only...(I guess). Lidar+photogrammetry would give us astonishing results.

I have to mention that HelloPhotogrammetry already gives me astonishing results comparing to alternative apps/services and expensive CG software. I've got clean and most detailed result I ever seen before (as a professional CG artist) just from 40+ photos. Industrial 3D scanners probably will do better but we have zero of them in our pockets and on our tables :)