Split screen: two speakers in one vertical frame
Who this is for
Interviews, podcasts, debates — two people talking, and a vertical feed where they do not fit side by side. Cropping to one face costs you the other person's reaction, which is often the whole point.
What the split does
The frame is divided horizontally into rows, one face per row. The viewer sees both, the faces stay large, the conversation reads.
There can be up to 4 rows. There is no next one, and not out of modesty: at 1920 pixels high it leaves a band where a face no longer reads — the technique stops doing the thing it exists for.
The split is not always chosen. It needs several speakers on screen and faces large enough to be worth showing as rows; with one face the frame simply follows whoever is talking.
The rows do not swap places during the clip. Who goes on top is decided once for the whole clip, by who speaks longer. The other stays at the bottom even when it is their turn to talk: faces jumping between rows read as an assembly error, even when every single swap is correct.
Why the seam has to be found
Here is the part nobody writes about.
Two-camera footage arrives already stitched: one camera on the left, one on the right. It seems obvious that the join sits exactly in the middle. It does not. In our own live order the seam ran at pixel 843 of 1920 — the cameras were combined off-centre, and nothing in the frame announces that.
Split it blindly down the middle and a strip of the neighbouring half rides up into the top row: 117 pixels of someone else's frame, stretched to 132. On screen it looks like a narrow vertical band on the right where you can see the second speaker's shoulder. We found this on our own clips and fixed it: the seam is now asked of the frame rather than computed from the width.
The test for a seam is strict, and that matters more than it sounds:
- it must be sharper than an ordinary spot in its own band of the frame — 5 times — not sharper than the frame overall: a high-contrast shoulder edge can be four times sharper than its neighbours and would pass as a seam;
- it must run the full height: the frame is checked in bands and the worst one counts;
- it must be visible on its own terms: a step of at least 2 on the 0-255 brightness scale. On a flat background "three times above zero" is still zero, and any edge would become a seam.
No seam found — we split down the middle, as before. The worst case stays exactly what it was, and that is deliberate: a false seam would cut a whole frame in two, which is worse than a strip.
What this means for your footage
- two-camera recordings work, even when the join is off-centre — the seam is found on its own;
- one whole frame with two people in a room works too: there is no seam, and the frame is divided the usual way;
- one thing worth checking after cutting: whether the top row has a narrow strip along its side showing part of another frame. It is the one split-screen fault that does not announce itself — the eye reads the face in the middle and skips the edge.
Where the subtitles go
In a two-row frame the text sits exactly where it does in a normal one: above the bottom zone the platform covers with its own interface. Those zones are covered in a separate article.