How face detection handles a crowd
Detection runs across the whole frame rather than on a single dominant subject, so large group shots are found in full. Each detected face becomes its own target you can assign to, skip, or leave alone.
One run swaps up to 10 faces — each is a separate pass, so the cap stops a single job fanning out indefinitely. In practice you hit a different limit first: pixels. Faces are found reliably while they still occupy a reasonable share of the frame, and because images are normalized to a 1920px long edge, someone standing at the back of a wide group shot may be resolved into too few pixels to swap convincingly even when they are detected. Crop tighter on the people who matter and the results improve markedly.