Change the Aspect Ratio in the Middle of a Video

Try the Mid-Video Aspect Shift Free

50 credits on signup, no card. Upload real footage and name the moment.

How to Change the Aspect Ratio Mid-Video

1. 1
   Pick your delivery canvas first

   Upload your footage (MP4, MOV, MKV or WebM, up to 14 GB or 3 hours) and set the shape you are publishing in — "make this 9:16 for TikTok" or leave it 16:9 for YouTube. This is the file's real aspect ratio, and it is fixed for the whole export. Everything after this happens inside it.
2. 2
   Name the moment and the shape

   Type the literal sentence: "go 9:16 when he picks up the phone, then back to full width when he puts it down". The agent places one shift at the first moment and a second shift back to 'source' at the second, using the word-level transcript and the frame tiles it reads to find them.
3. 3
   Tune the morph in plain English

   "Make the close-in slower — take two seconds." "Make the bars white." "Don't push the picture in when it narrows." The morph runs anywhere from 0.1 to 4 seconds, the bars take any hex colour, and the push-in is optional.
4. 4
   Preview, then export

   The preview renders from a proxy so you can watch the change immediately; the agent looks at the frames it produced to check the shift landed where you meant. When it is right, export — the final render comes from your original file at source quality.

The one-time analysis runs once per upload. Placing, moving or removing an aspect shift after that is a single chat message and re-renders in seconds — it changes no timing, so nothing downstream has to be recomputed.

What Actually Happens to the Frame

The canvas never changes size. What moves is a pair of matte bars sliding in from off-screen, and the picture pushing in behind them so the subject holds its size instead of shrinking with the window.

How the Shift Works Underneath

A rendered file has one resolution for its entire duration. There is no container that renegotiates its dimensions halfway through, and no player that would follow it if there were. So a mid-video aspect change is an animated matte inside a fixed canvas — and once you accept that, it stops being a compromise and becomes the reason the effect is safe.

The target shape is resolved as a fraction of the canvas. A 9:16 window inside a 16:9 canvas is narrower than the canvas, so it takes the full height and 32% of the width — pillars. A 16:9 window inside a 9:16 canvas is wider, so it takes the full width and 32% of the height — a letterbox. 1:1 inside 16:9 keeps 56% of the width. The bars are half of whatever the target gives up on that axis, on each side, so the window is always centred on the canvas.

Each bar is a full-frame colour plate slid in from off-screen, and the visible part of it is the bar — its thickness animates because its position does. That indirection is not decoration. The obvious filter for the job evaluates its geometry once, when the render is configured, so a bar whose width is written as an expression in time renders at a constant width for the whole clip; the compositing overlay re-evaluates per frame, which is what makes a morph possible at all. The animation is eased in and out over the duration you asked for, so the frame accelerates away from its old shape and settles into the new one rather than sliding linearly.

The push-in is proportional to what the frame gave up. A shift that costs 68% of the frame area pushes the picture in by 34%; a 1:1 shift, which costs 44%, pushes in by 22%. The gain is capped at 35% — beyond that a subject's head starts leaving the top of a vertical crop, which is worse than a subject that shrank slightly. It rides the same geometry pass the zooms use, so a punch-in over the same moment composes with the shift instead of replacing it, and the aim of that zoom applies to the close-in too.

The bars are composited above captions and above the on-screen text layer, deliberately. When the frame narrows to 9:16, anything outside the new window is outside the frame — and a caption half-hanging into the letterbox is the single artefact that would give the whole effect away as a mask rather than a reframe.

Because it is a matte, it touches nothing else in the edit. No segment is retimed, no audio is resampled, no caption event moves by a frame. That is the invariant that lets you place, move and remove aspect shifts freely late in an edit, which is the point in a project where most effects have become expensive to change.

How It Survives Later Cuts

"Go vertical when he says the thing" belongs to the moment he says it, not to a clock reading. So each shift is stored against the footage underneath it, and when you cut something earlier the shift is re-anchored through the new timeline and the agent tells you where it moved to. If the moment it was placed on is cut away entirely, the shift is removed rather than left behind, and you are told that too.

The alternative — leaving it at a fixed output timestamp — is the version that fails silently. Unlike a sound effect drifting a second late, a drifting matte is a full-screen change over the wrong sentence, and nobody could mistake it for anything but a bug. This is the same reasoning behind Valmera's honesty layer generally: every agent reply is verified server-side against the edits actually recorded, so the agent cannot report a shift it did not place.

How Platforms Handle a Changing Frame

A player reads the file's dimensions once and sizes itself to them. It does not resize during playback, on any platform. This has one practical consequence that decides the whole edit, and it is worth getting right before you place a single shift: pick the delivery canvas first, then shift inside it.

When It Goes Wrong

The subject leaves the window. The window is centred on the canvas. A speaker standing on the left of a 16:9 frame is simply not inside a centred 9:16 crop of it. The fix is to aim a zoom at them over the same stretch — "push in on him while it's vertical" — because the shift and the zoom share one geometry pass, so the aim carries.

The picture looks soft while the frame is closed. The push-in magnifies, and on footage already at output resolution a 34% push is a 34% magnification. If the softness reads on your material, say "don't push in when it goes vertical" and take the shrinking subject instead.

Two shifts too close together. Morphs cannot overlap — two of them animating inside each other would have the bars travelling to two places at once, with the result depending on evaluation order nobody can read off the edit. So a shift placed inside another's morph is refused with the id and timing of the one it collides with. Move it later or shorten the first.

Shifting to the shape it is already in. This is refused rather than written. It would render nothing visible, and an agent reporting a change that exists nowhere on screen is worse than an agent that says no. To come back out of a shape, ask for 'source' — "open back out to full width" — not for the canvas ratio by name.

Captions clipped by a narrow window. Caption style and position apply to the whole edit rather than per section, so if the frame closes to 9:16 anywhere, size and position the captions to fit the narrowest window in the edit — smaller, centred, and inside the safe strip. See captions for the size and position controls.

How People Do This Without Valmera

It is doable in every serious editor, and it has never been quick in any of them, because the aspect ratio is a property of the sequence and the sequence is not something the timeline expects you to animate.

Premiere Pro. The frame size belongs to the sequence, and it is one size for the sequence's whole length. Auto Reframe — whether you run it on a sequence or apply it as an effect to a clip — repositions the content inside that frame; it is not a way to change the frame's shape partway through. For a mid-video change you keep the sequence in your delivery shape and animate the shape on top: a Crop effect on an adjustment layer with keyframes on Left and Right (or Top and Bottom), plus a keyframed Scale on the footage below so the subject holds size. That is a keyframe pair on each cropped edge and another on Scale, per shift, placed by hand at every moment you want the frame to move.

DaVinci Resolve. The usual route is a Fusion composite or an adjustment clip carrying a rectangle mask with keyframed width, with the Edit page's Sizing zoom keyframed to match. Resolve gives you the most control of the three and asks for the most setup; the timeline's own resolution is still a setting on the timeline, fixed for its whole length.

ffmpeg. Reachable, and instructive about why the effect is fiddly. A drawbox with a time-varying width will not animate — its geometry is evaluated once at configuration — so you end up compositing a colour source with an overlay whose position is an expression re-evaluated per frame, then adding a zoompan for the push-in and easing both curves by hand. That is the same construction Valmera emits; the difference is that you are writing the filtergraph, and every timing change means recomputing every expression.

Browser tools and mobile apps. CapCut, Veed, Kapwing, Clipchamp and Canva all resize the whole project — pick 9:16, everything becomes 9:16. Some let you keyframe scale and position on a clip, which gets you a partial version of the look if you split the clip at the moment and animate manually. In each of them the aspect ratio is a property of the project canvas rather than something you place on the timeline, so a mid-video change is work you assemble yourself — which is the gap this page exists to describe.

Honest Limits

• The export's resolution is fixed. A shift changes the visible frame, never the file's dimensions — no platform will show a shifted section full-screen.
• The window is centred on the canvas. Off-centre subjects need a zoom aimed at them over the same stretch.
• Six targets only: 16:9, 9:16, 1:1, 4:5, 4:3 and source. No arbitrary ratios.
• Morph length is 0.1–4 seconds, and two morphs cannot overlap.
• The push-in is capped at 35% and is proportional to the frame area lost — you can disable it, but not raise it past the cap.
• Bars are one flat colour for the whole edit; there is no blurred-backdrop option for a mid-video shift, though whole-video resizing does offer blurred pad.
• Captions are styled and positioned once for the whole edit, so a narrow window may hide text placed for the wide one.

Related Tools

A shifting frame is usually one beat inside a larger edit. The same chat can set the delivery aspect ratio for the whole export, aim zooms and punch-ins that compose with the shift, add word-accurate captions sized for the narrowest frame, cut the dead air before you place the shifts, and style the cuts around them. If you want the numbers behind a ratio, the aspect ratio calculator works them out.

Every editing tool Valmera has — 97 of them, plus 11 session tools — is also published as a Model Context Protocol server, so Claude can place an aspect shift on real footage from inside a conversation. Browse the full set in the tools hub or read the reframing docs.

Put the Frame to Work

Describe the moment. Watch the frame move. Free plan, no card.

Frequently Asked Questions

Change the Aspect Ratio Mid-Video

The agentic AI video editor — describe the edit, review the preview, export from your original file at source quality.