Prompt what the model looks for in every frame

This is a short clip from a fixed overhead camera in a print shop. In the top of the view stands a small hand-fed printing press. At the lower left, on an orange table, is a stack of flat unprinted paper bags waiting to be printed. The operator of the press sits directly under the camera: when he is at his place you see his hands and forearms, and sometimes the top of his head, at the bottom of the frame. A pair of green gloves often lies on the table. The frames run in time order. Look at every frame, one by one. Return one JSON object with these fields: - evidence: In at most 20 words, say what you see of the operator across the clip. - operator_at_the_workstation: True if any frame shows a hand, bare or gloved forearm, shoulder, or top of a head at the bottom. Green gloves lying flat, tools, paper, and folders do not count. False only if no frame shows any part of a person. - unprinted_bags_left: True if flat unprinted paper bags are still stacked at the lower left.

Reference images optional

＋ Add reference image

Source video up to 30 minutes