How to Convert a Horizontal Video to a Vertical Clip That Follows the Speaker
A 9:16 crop keeps less than a third of a 16:9 frame, so the whole job is deciding which third. Here is how a subject-centered crop finds the speaker and follows the conversation, when a square window on a blurred canvas is the better choice, and how clippers run either one on every clip without keyframing by hand.
To convert a horizontal video to vertical, cut a 9:16 window out of the 16:9 frame and keep the person talking inside it. That window is only about a third of the frame's width, so a fixed center crop loses whoever sits off center. Use a subject-centered crop for one speaker, and a square window on a blurred 9:16 canvas for a two-shot.
Why a straight crop goes wrong
A 16:9 frame at 1920x1080 is wide. A 9:16 window the full height of that frame is 608 pixels wide, which is 32% of the picture. Two thirds of the frame is gone whatever you do, so the only question is which third you keep.
A centered crop keeps the geometric middle. That works for a single person framed in the center of a talk or a vlog. It fails on almost everything clippers are paid to cut:
- Podcasts shot on one wide camera. The host sits left, the guest sits right, and the middle third is a microphone and a table.
- Interview setups. A guest framed on the rule of thirds is half out of a centered window.
- Multicam shows. The edit cuts between cameras, and each camera frames its person somewhere different. A crop aimed for the first shot is wrong for the next.
- Streams and gameplay. The face cam sits in a corner, nowhere near the middle.
Doing it by hand means setting a crop for each shot and keyframing a new position every time the speaker changes. Fine for one clip, slow for a day of them.
The four ways to make a horizontal video vertical
There are really four choices, and each one suits a kind of shot.
1. Crop to 9:16, centered. The middle third, full height. Right when the subject is centered and alone. Fast and predictable, and wrong as soon as anyone moves off center.
2. Crop to 9:16, subject-centered. The same window, aimed at the person. This is the default for talking clips because it fills the vertical frame with a face, which is what the feed rewards. The catch is that a narrow window can only hold one person at a time.
3. Square on a blurred 9:16 canvas. A square window out of the frame, 56% of a 16:9 frame's width instead of 32%, sits in the middle of a 9:16 canvas filled with a blurred, darkened copy of the same picture. You keep much more of the scene, so a two-shot can stay a two-shot. The blurred bands above and below the square give captions a clean place to sit.
4. Full frame on a blurred 9:16 canvas. The whole 16:9 picture, scaled to the canvas width, over the same blurred fill. Nothing is cut out of the frame, but the picture is small. Use it when the whole frame matters: a wide demo, a slide, a scene where the action spans the shot.
Plain letterboxing (the frame on black bars) is the fifth option. It keeps everything and looks like a repost, which is the opposite of what a clip should look like on a vertical feed.
How to pick, shot by shot
Look at the source for three seconds and you can usually tell which mode it needs.
| What the shot looks like | Use this | Why |
|---|---|---|
| One person talking to camera | Crop 9:16, subject-centered | Fills the frame with the speaker |
| Multicam podcast, one person per shot | Crop 9:16, subject-centered | The window re-aims at each camera cut |
| One wide shot of two people talking | Square on blurred 9:16, or subject-centered if you want the crop to follow the talker | Square keeps both; the crop jumps between them |
| Panel or video call with three faces | Crop 9:16, subject-centered | Follows whoever is speaking |
| Screen share, slides, gameplay with HUD | Full frame on blurred 9:16 | Text at the edges stays readable |
Whatever the mode, put captions where nothing important is, and scale the result to 1080x1920 before you post: a crop out of a 720p source is only 404 pixels wide, and captions burned at that size look soft.
How a subject-centered crop follows the speaker
"Find the face and center on it" sounds simple and breaks in a few predictable ways. Here is how Extract Clip on Treza handles them, since that is the part people ask about most.
- It finds people in sampled frames. A vision model looks at frames sampled across the clip and anchors the vertical window on the speaker.
- It splits the clip at camera cuts. On a multicam podcast every camera sees the same room and lighting, so cuts are subtle. Extract Clip detects them and gives each shot its own anchor, so the window re-aims when the edit cuts from the guest to the host.
- It follows whoever is talking. A still frame shows where each person is, not who is speaking. In shots that hold several people, a model that hears the clip works out who is talking when, and the window cuts to that person at each turn. During silence it holds on the last speaker.
- It aims at the subject when there is no face. A profile shot, an animal or an object on screen has no face to detect, so those shots are aimed at the main subject of the frame instead of falling back to the middle.
- It falls back safely. If detection finds nothing usable, the clip uses a centered crop rather than failing, and the run log says so.
The window moves in cuts, not pans, the way an editor would reframe a conversation by hand.
Doing it on every clip, not just one
On a campaign you might cut a week of clips from one source, so the reframe should be a setting, not a session. On Treza, reframing is one setting on the Extract Clip node inside an AI video pipeline:
- Source. Drop the source file on a File node or paste a direct MP4 link. A Media Download node can pull a video from a Vimeo, X or Twitch page URL. Download only what the campaign or the rights holder allows you to clip.
- Range. Type a start and end, or wire in a model that picks the moment from the transcript.
- Reframe. Choose one of the modes above: crop to 9:16, 4:5 or 1:1 (centered or subject-centered), pad to 9:16, square on a blurred 9:16 (centered or subject-centered), or full frame on blurred 9:16. Extract Clip reads long masters in place, so a two-hour recording is not re-rendered to make a forty-second clip.
- Captions. A Captions node re-transcribes the clip and burns word-accurate captions in, with an optional spoken-word highlight and a color per speaker. Set the position to Bottom on a canvas mode so the words sit in the blurred band. Our guide to adding captions automatically covers the caption side in more depth.
- Post. Treza posts to YouTube Shorts and, with a TikTok Upload node, to TikTok. For Instagram Reels and X, the pipeline hands you the MP4 and you upload it by hand.
If you clip for paid campaigns, the Campaign Clip template already has this wired, with the subject-centered 9:16 crop as its default. It reads the brief, picks the moment, cuts and reframes it, captions it, has a judge model score the finished clip against the brief, and posts it. The clipping guide explains how campaigns pay, and the Whop, Vyro and Clip Collective pages map the template onto each marketplace's workflow. Treza is independent of all of them.
No video generation model runs in any of this; it is cutting, reframing and captioning real footage. If you need original footage instead, that is the AI video generator.
Reframing and campaign rules
Campaign briefs set the format, and most ask for vertical clips. Read the brief before choosing a mode: some campaigns want the creator kept on screen, some forbid edits that change the meaning, and a letterboxed frame often reads as a straight repost. Marketplace terms also prohibit reposting the same clip on one account; Vyro's content requirements say so directly as of October 2026. A good reframe is part of what makes each clip your own edit, but it does not change what the campaign allows. If a rule is unclear, ask the campaign before you post, and see why clips get rejected for the rules that trip people up most.
One tradeoff to know
A subject-centered crop is only as good as the shot. A wide shot where two people talk over each other will cut back and forth often, and a crop cannot add pixels to a low-resolution source. If the result looks busy, switch to square on a blurred canvas.
Frequently Asked Questions
How do I convert a horizontal video to vertical without cutting people off?
Use a subject-centered crop, which aims the 9:16 window at the person instead of the middle of the frame. If two people must stay in shot together, put a square window on a blurred 9:16 canvas, which keeps 56% of the frame's width instead of 32%.
What is auto reframe in video editing?
Auto reframe is a crop that positions itself. Instead of you setting the window for each shot, software finds the subject in the frame and moves the window to keep them in view, re-aiming when the camera cuts or the speaker changes.
How do I change a 16:9 video to 9:16 for TikTok and Shorts?
Crop a 9:16 window out of the full height of the frame, place it on the subject, and scale the result to 1080x1920. Add captions after the reframe so they are positioned for the vertical frame, then post the MP4.
Should I crop or use a blurred background for vertical clips?
Crop when one person is talking, because a face filling the frame reads best on a phone. Use a blurred background when the shot needs more width, such as two people in one wide shot, slides, or gameplay, and put the captions in the blurred band.
Can the vertical crop follow whoever is talking in a podcast?
Yes. On Treza, Extract Clip's subject-centered crop splits the clip at camera cuts and, in shots with several people, uses a model that hears the audio to cut the window to whoever is speaking at each turn.
Does reframing make a clip count as original content on clipping campaigns?
Reframing is part of editing a clip, but whether a clip is acceptable is decided by the campaign's brief and the marketplace's terms. Clip only what the campaign allows, follow its rules, and never post the same clip twice on one account.
