Make Vertical Clips From a Horizontal Podcast Recording
To turn a horizontal 16:9 podcast into vertical 9:16 clips, you cut a tall window out of the wide frame and decide, second by second, which part of the picture it follows. One person on screen: center on the face. Two people trading lines: track whoever is talking, or stack them in a split-screen. The aspect-ratio change is the easy part, keeping the right face on screen when the conversation moves is the work.
Most editors get the crop and lose the reframe. They draw a tall box, park it in the middle of a two-shot, and ship a clip with two half-faces and a microphone wedged between them. The fix is to treat reframing as a per-scene choice, not a one-time setting. This guide gives you the decision rule for picking centered crop, speaker tracking, or split-screen, plus the source-framing test that tells you when split-screen is even possible, and where faces and captions go once you have chosen.
Why the reframe is the real blocker, not the crop
Cropping a frame is trivial. Any tool can draw a 9:16 box. What breaks is everything downstream: the active speaker drifts to the edge, a guest's reaction gets cut off, the camera switches and the box jumps to a blank wall. The crop is one decision. The reframe is hundreds of them, one per moment the talking changes hands.
This matters because vertical clips are how shows get found now. Clips drive 20–40% of new audience for video podcasts and can raise reach 2–5×, a single-studio client range, directional rather than audited (Podcast Studio Glasgow). The surface they live on is a phone held upright, a feed where the face has to read at a glance or the viewer scrolls. A reframe that loses the speaker turns a strong moment into a skip, the case for keeping the action in frame when you reframe.
The reframing decision rule: reframe by who is talking
Pick the reframe from the scene, not the source footage. The wide shot does not tell you what to do, the conversation does. Run every clip through three questions before you touch a crop tool.
- How many people are on screen and speaking? One speaker is a centered crop. Two trading lines is a tracking or split-screen call.
- Does the moment depend on a reaction? If the payoff is the other person's face, a laugh, a flinch, a "wait, what", both faces have to stay visible. That rules out hard tracking.
- Does the camera cut during the clip? A single static camera is easy. Multi-cam footage needs the reframe re-checked at every switch, because the speaker's position in the frame changes the instant the angle does.
Those three answers map cleanly to three reframes. Here is the rule as a table you can keep open while you edit.
| Scene | Use this reframe | Because |
|---|---|---|
| One person, single camera | Centered crop on the face | Nothing to track; lock it and move on |
| Two people, fast back-and-forth | Active-speaker tracking | Keeps the talker filling the frame |
| Two people, reaction-dependent | Split-screen (stacked) | Both faces stay on screen for the payoff |
| Multi-cam, switches mid-clip | Re-frame per angle | The speaker's position jumps at every cut |
How to do each reframe
One person: centered crop
Put the face in the upper-middle of the 9:16 frame, eyes around the top third, not dead center. A face centered vertically leaves a forehead-heavy top and wasted floor at the bottom. Lift it. For a solo host who leans and gestures, leave a little headroom so a lean does not clip the top of the head. This is the only reframe you can mostly set and forget.
Two people: active-speaker tracking
Tracking moves the crop window to whoever is talking. When it works, the talker fills the frame and the clip cuts like a conversation built for a phone. When it fails, the window whips back and forth on every "yeah" and "mm-hmm" and makes viewers seasick. Add a hold: ignore interjections under about a second so the window only moves on a real handoff. If your footage is two people seated far apart, tracking is the right call, a centered crop would jam both of them into the edges with the table in the middle. That tiny-floating-heads failure is the most common reframing mistake, and there is a dedicated fix in why vertical crops cut off faces and how to stop it.
Two people: split-screen
When the reaction is the point, stack both faces, one in the top half of the 9:16 frame, one in the bottom. The viewer sees the setup and the response in a single shot, which is exactly what tracking throws away. Split-screen costs you size: each face now lives in a roughly 9:8 box instead of the full frame, so it only works when both people are framed tight in the source. The four ways to fit two heads in one tall frame, and which conversation each suits, are laid out in framing two speakers in one vertical clip.
The source-framing test for split-screen
Before you commit to split-screen, check the source. Split-screen only works if both faces are already shot tight enough to survive being shrunk to half a tall frame. If one camera is a wide room shot, stacking turns that person into a distant dot the moment you cut them to half-height. Run this quick test on the wide frame: imagine each person's head scaled down to fit a box that is half the height of a phone. If a face would end up smaller than a thumbnail at that size, split-screen is off the table for that clip, fall back to tracking. This is the check most reframing guides skip, and it is why split-screens look great on close two-shots and terrible on lecture-hall footage.
Multi-cam: re-frame at every switch
Multi-cam is where reframing quietly breaks. A program feed cuts between a wide two-shot, a close-up on host A, and a close-up on host B. One crop setting cannot serve all three: the box that frames host A's close-up will land on empty space the instant the feed cuts to the wide. Treat each camera angle as its own reframe. Most AI clippers re-detect the speaker at a cut, but imperfectly, which is why this is the angle most worth reviewing by hand.
Use this checklist on every multi-cam clip:
- Mark each camera switch in the clip first, before reframing, they are your reframe boundaries.
- Confirm the active speaker is centered in the new angle right after each cut, not the person who was talking in the last shot.
- Watch the wide shots specifically, they are where a single crop most often lands on a wall or an empty chair.
- Check the first frame after every switch, because a one-frame miss reads as a glitch on a phone.
If your auto-reframe keeps landing on the wrong spot after a switch, the diagnosis and fixes are in why auto-reframe keeps cutting to the wrong thing.
Where the face goes, and where captions go
A 9:16 frame is not one canvas; it is three bands. The top is the hook zone where on-screen text earns the first second. The middle is the face-safe zone. The bottom is contested, captions want it, and so does every platform's UI.
Keep the face in the central band and pin captions just below it, never at the very bottom edge. TikTok's right-side action rail and bottom caption/handle area, plus Instagram's lower UI, eat the bottom roughly 15–20% of the frame; anything you place there gets covered on at least one platform. The full safe-zone map by platform is in caption placement and safe zones in vertical video. Captions are not optional either way: most social video is watched on mute, with about 75% of mobile video watched without sound (Verizon Media / Sharethrough, 2017, via Digiday), and an older Digiday (2016) figure putting around 85% of Facebook video as silent (publisher-reported, directional). No captions, no clip.
Common mistakes (and the fix)
- Center-cropping every two-shot. It produces two half-faces and a mic. Fix: switch to tracking or split-screen the moment two people share the frame.
- Forcing split-screen on wide footage. A distant speaker stays distant at half-height. Fix: run the source-framing test first; if a face would be thumbnail-small, track instead.
- Letting tracking twitch on backchannel. The window jumps on every "yeah." Fix: add a ~1-second hold so it only moves on a real speaker handoff.
- One crop for multi-cam footage. The box that frames the close-up lands on a wall in the wide. Fix: re-frame per camera angle and check the first frame after every switch.
- Captions at the very bottom. They vanish under the platform UI. Fix: lift the caption band above the bottom ~15–20% on every platform.
- Reframing before you trim. You waste effort fitting frames you will cut. Fix: lock the in/out points first, then reframe what survives.
How AI auto-reframe fits in
Auto-reframe does the mechanical work, it detects faces, finds the active speaker, and moves the crop so you are not keyframing a box by hand for twenty clips. That is the time saver, and it is real. What it does not do is judge intent. It cannot tell that the funny moment is the listener's face, not the talker's, so on reaction beats it will track to the wrong person. It re-detects at camera switches, but not always correctly.
The honest framing: every AI clipper still needs roughly 20–40% human review, and reframing is where most of that review goes. Treat auto-reframe as a fast first pass, the speaker-tracking detail behind it is covered in how AI clip detection actually works, then scan each clip once for the two failure modes that matter: a lost active speaker and a buried caption. That review is minutes, not hours, and it is the difference between a clip that reads and a clip that gets scrolled.
FAQ
Can you turn any horizontal podcast into vertical clips? Yes, if there is usable video. A 16:9 recording reframes to 9:16 by cropping a tall window from the wide frame. Audio-only episodes have no video to crop, those need a generated visual like an audiogram instead. The reframe quality depends on how tightly the source was shot: wide room footage gives you less to work with than a close two-shot.
Should I use speaker tracking or split-screen for an interview? Use tracking when the conversation is a fast back-and-forth and the talker's face carries the moment. Use split-screen when the payoff is a reaction, a laugh or a stunned look, because tracking would cut away from the face that makes the clip land. Match it to the moment, not to the whole episode.
Why does my auto-reframe keep landing on the wrong person? It is tracking the loudest or most-detected face, not the one that matters to the moment, common on reaction beats and right after a camera switch. Re-check the crop at every multi-cam cut and override the reframe on reaction shots. The full diagnosis is in why auto-reframe keeps cutting to the wrong thing.
Where do captions go in a vertical podcast clip? Just below the face, in the lower-middle band, never at the very bottom edge, where the platform UI covers them. Keep nothing important in the bottom ~15–20% of the frame, since TikTok's action rail and Instagram's lower UI sit there.
Does reframing hurt video quality? Cropping discards pixels, so a tall slice of a 1080p-wide frame has fewer pixels than the original. If the source is 4K or a tight shot, the loss is invisible. If it is 720p and wide, the cropped face can look soft. Shoot tighter or higher-resolution when you can, and keep both faces close in the source to start with.