Landscape footage, converted to 9:16 without losing the subject

The arithmetic of going from 16:9 to 9:16

Take a standard 1920 × 1080 frame. To fill a 9:16 output using the full height of that frame, you can keep a column roughly 608 pixels wide. That is a little under a third of the original width, and the other two thirds are discarded outright. Whatever was in them is not recoverable, compressed, or shrunk — it is gone.

There is a second consequence people notice later. That 608-pixel-wide column then has to be scaled up to 1080 pixels wide to hit a standard vertical output, so you are enlarging what remains by roughly 1.8 times. On a 1080p source that is a genuine upscale and it is visible on close-ups. If you have a 4K master of the same footage, use it — the same crop takes about 1216 pixels of real width out of a 3840-wide frame, which is more than enough to fill 1080 without stretching anything.

None of that is a criticism of vertical video, it is just the physics of the format. It does explain two practical rules. Convert from the highest-resolution source you have rather than from a compressed re-upload, and expect that a conversion decision is really a decision about which third of your picture is worth keeping.

What leaves the frame, and the mistakes that follow

The obvious loss is people. A guest sitting at the right of a two-shot is entirely outside a centred vertical crop, and unless the crop moves, the clip becomes a video of the host reacting to an unseen voice. Tracking exists for this and it is the reason the same footage can produce a usable vertical clip at all.

The less obvious loss is graphics. Lower thirds, name captions, logos, chart annotations and slide edges tend to live near the borders of a landscape frame, precisely because that is where they stay out of the way. Crop to vertical and they are the first casualties. If your footage depends on on-screen text, either use fit so the whole frame survives, or accept that the text needs recreating at vertical dimensions.

Then there is composition. A wide shot that was framed to show a room, a stage, a landscape or a group is not the same shot once two thirds of it is removed — it is a detail from that shot. Sometimes the detail is better for social. Often it is not, and the honest move is to recognise that a particular piece of footage was made for a horizontal frame and leave it there.

Three conversion strategies, and a workflow for picking one

The first is scale-to-fit with padding: shrink the whole landscape frame, keep it intact, and fill the space above and below with black or a blurred duplicate. It loses no information and it is the least effort. Its cost is that your subject ends up small in a format viewed on a phone at arm's length, and the blurred-bar look reads as a re-upload rather than something made for the feed. Use it when the frame genuinely cannot be cut.

The second is a fixed crop, usually centred. It fills the screen properly and requires no analysis, so almost every free converter offers exactly this and nothing else. It works on a single, centred, stationary subject and fails on essentially everything else, quietly, in a way you only notice after uploading.

The third is a tracked crop, which is what this tool does by default. The crop window is placed where the subject is and moves as the subject and the shot change, with a switch to split or grid layouts when there is more than one person worth showing. It fills the frame like a fixed crop and keeps the subject like fit does. It costs analysis time, which is why it belongs in an automated pipeline rather than in a manual export dialog.

Choosing between them takes about ten seconds if you ask the questions in the right order. Is there text or a graphic near the edge of the frame that a viewer needs to read? If yes, fit. Is there more than one person who speaks during the section? If yes, split or a grid. Is there a single subject who moves, or a camera that reframes them? If yes, a tracked crop. Only if the answer to all three is no does a plain centred crop become the right and cheapest choice.

Troubleshooting a conversion after it has already rendered

Most people read a bad vertical export as a failure of the analysis. Far more often the analysis was right about the footage and wrong about your intent, and the correction is one setting rather than a re-shoot. Four symptoms cover nearly all of it, and each has a different cause.

The subject pinned to one edge of the tall frame for a long stretch means the crop locked onto a face that is not the one you care about — a listener nodding, a person in the background, a poster with a face on it. Switching that clip to the split layout resolves it when there really are two people, and forcing a wider static layout resolves it when there are not.

A missing lower third, a vanished logo or a chart with its axis labels sliced off is not a bug at all. That is a crop doing the only thing a crop can do, and the graphic was living near the border of the wide frame precisely because that is where it stayed out of the way. Re-render that clip as fit and the whole original picture survives at a smaller size.

Soft, mushy close-ups are almost always an input problem rather than an output one. A 1080p file downloaded back off a platform has already been through one round of compression, and the vertical crop then enlarges those compressed pixels by roughly 1.8 times. In practice the fix lives upstream: convert from the master file on your drive, not from the published copy.

And a clip whose framing simply looks restless — never wrong, never still — usually contains a speaker who rocks in and out of shot. Tracking is doing its job; the job is the wrong one for that footage. A fixed wider composition looks calmer even though it makes the person smaller, and calm beats large at this size more often than people expect.

When converting to vertical is not the right move

The clearest case is a shot whose subject is the width itself. A landscape, a stage with a band spread across it, a football pitch, a two-page diagram — remove two thirds of the frame and what is left is not a tighter version of that shot, it is a different and worse shot. Fit will preserve it, but preserving a wide composition inside a tall frame at one third the size is rarely worth posting either.

Worth knowing before you batch a whole archive: footage that carries its meaning in on-screen text almost never converts well. Slide decks captured at desktop scale, spreadsheets, code editors and dashboards are legible on a laptop and illegible on a phone at any aspect ratio. The honest answer there is to rebuild the visual for vertical rather than to convert the recording of it.

The other mismatch is scope. This produces clip-length verticals from the parts of a recording that stand alone; it does not re-render an entire session end to end in a tall frame. Say you need all ninety minutes of a conference talk available vertically for an in-app player — that is a straight transcode with a fixed crop, and an editor or a single ffmpeg command does it faster and cheaper than any analysis pipeline will.

Finally, if the source was already shot vertically, the conversion half of this does nothing for you. The moment selection, the sentence-boundary cutting, the filler removal and the captions still apply, but you should judge the tool on those rather than on framing you never needed.

Doing the reframe by hand versus letting the footage decide

Manual reframing in an editor is not hard, it is just repetitive. You drop the landscape clip into a vertical sequence, scale it up, and keyframe its horizontal position so the crop sits on whoever is speaking. On a single clip with two or three speaker changes that is a few minutes of work and the result is exactly what you intended, frame for frame, which no automatic pass can promise.

The arithmetic changes with volume, and the biggest mistake people make is pricing the first clip instead of the twelfth. A one-hour recording that yields twelve postable moments carries perhaps thirty speaker changes across those twelve clips, and each one is a keyframe pair you set, watch back, and nudge. That is the afternoon, and it is why hand reframing quietly stops happening about three weeks into anyone's clipping habit.

The split is therefore not automation versus craft, it is where you spend the craft. Let the analysis take the routine batch, then open the two or three clips that carry real weight — the ad, the launch video, the one going on the homepage — and reframe those yourself with the crop exactly where you want it. Both paths export the same 1080 × 1920 file, so nothing downstream cares which route a clip took.

One genuine advantage of the manual route is that a human knows what the shot is about. An automatic pass follows faces and shot type; it does not know that the product on the table matters more than the presenter this time. When the meaning of a shot sits somewhere other than on a face, hand framing wins and it is not close.

Alternatives to an automatic converter, and what each costs

The alternative nobody counts as one is filming for the frame you will publish in. A phone on a second tripod, shooting vertical alongside the main camera, produces material that needs no conversion at all and no upscale. It costs a device and a bit of setup discipline, and for anything you record regularly it is the highest-quality option available. Conversion exists for the footage that already happened.

Deliberate letterboxing is a real alternative and gets dismissed too quickly. Some formats — a chart walkthrough, a wide product shot, an archive clip — are more honest presented whole with framing above and below than cropped into a detail. It reads as a repost when it is lazy and reads as a choice when the content plainly needs the width, and viewers can tell the difference.

The in-app editors on TikTok, Reels and Shorts will all trim and reposition a landscape file for free, on the phone, in the time it takes to upload. For one clip whose good part you already know, that is the correct tool and any other answer is overengineering. It falls apart at the point where you have to find the good parts of an hour of footage before you can trim anything.

Command-line and editor routes are covered in the comparison below, and both remain the right call when you already know the exact crop you want. The last option is other clipping services, most of which now do face-aware conversion of uploaded files. If you also work from live broadcasts, that is the axis worth testing them on — clipping during a stream is a different engineering problem, and the livestream clip generator explains why the two behave differently.