why not skip the text conversion? is it usable at all?

gemini embedding 2 converts straight video to vectors. in this case, dashcam clips don't have audio to transcribe and even if they did, it would be useless in the search