Which shape should the picture-in-picture window be?

What each shape keeps, and what it throws away

The tall rounded rectangle matches the proportions of the viewfinder itself, so it is the shape that crops the least from the source camera. The square and the circle are both true 1:1 crops of that same frame, taken from the centre, and they discard more of it than the rectangle does regardless of which size step you pick. A circle is masked further inside that square crop, so a face or hand near the edge of the square can end up clipped by the round outline even though the pixels underneath were captured — the crop happened at the square, the circle just hides the corners of it on top.

This matters most for anything where the edges of the small window carry information: a hand reaching into frame, a second person leaning in, a gesture that starts inside the window and finishes outside it. The tall shape gives that kind of movement the most room before it runs into an edge. A circle gives it the least.

Tall: reads as a continuation of the shot

Because it shares the canvas's own proportions, a tall inset does not read as a separate object dropped onto the frame the way a square or circle does — it reads more like a second, smaller version of the same kind of shot. That makes it the right call whenever the inset needs to be watched for more than a face: hands doing something, upper-body movement, a second person who might lean into or out of frame during the take. It is also the shape that keeps the most resolution from the source camera, which matters more at 1080p than at 720p.

Square: a neutral, self-contained window

A square inset reads immediately as its own distinct element rather than a continuation of the main shot — closer to a thumbnail than a second camera angle. That is a reasonable default when the small window is meant to sit quietly in a corner without competing for attention: a location shot, a static product, anything the viewer is not meant to study closely. It crops more than the tall shape but keeps a squarer, more even margin around whatever is centred in it than a circle does.

Circle: reads as a reaction, on sight

A circular inset is the shape most associated, by convention across video generally, with "this is a person reacting to something" rather than "this is a second thing happening." A viewer registers that meaning almost instantly, before they have consciously read the content inside it, which is exactly why the shape works well for reaction and commentary framing where the inset is a face and only a face. It works less well the moment the inset needs to show anything beyond a fairly centred head and shoulders — the round mask is the least forgiving of the three shapes for content that drifts toward the edge.

None of this has to be decided before you press record

Shape, like corner, size and layout, can be changed in DualCam while a recording is already running, and the file follows the change immediately because the preview and the recording are the same rendered frame. The only settings locked the moment recording starts are resolution, mute and the countdown timer. That means the practical approach is to start with whatever shape fits how the take opens, and switch if the small window turns out to be doing a different job than you expected once you are a minute into it.