How do I change a video's aspect ratio without cropping it?

You have three options. Padding (letterboxing) keeps every pixel but shrinks the video inside bars. AI outpainting, offered by tools like Runway Expand Video (2024) and Luma Dream Machine Reframe (2025), keeps your original frame intact and generates new content at the edges; the fill looks plausible but is hallucinated. Re-rendering regenerates the video natively in the target ratio with every shot recomposed, so nothing is discarded and nothing is patched onto the edges. Re-rendering applies to AI-generated video, which is how Riffkit reframes a 9:16 winner into 4:5, 1:1, or 3:4.

What is video outpainting?

Video outpainting is a generative AI technique that extends a video beyond its original borders. The model keeps the original frame untouched and synthesizes new content around it until the target aspect ratio is filled. Runway Expand Video, Luma Dream Machine Reframe, Kling Extend, and Pika Expand Canvas all work this way. The strength is that no original pixels are lost; the weakness is that everything added at the edges is invented by the model, which can introduce objects that were never in the scene and leave visible style seams.

Is cropping or outpainting better for ad creative?

Neither is ideal when your composition carries information. Cropping discards content near the edges, so captions, products, and secondary subjects can get cut out. Outpainting keeps your footage but generates hallucinated content next to it, which is risky beside a real product. For a simple centered talking head, cropping is fine and cheap. For ad creative with on-screen text, product shots, or multiple subjects, re-rendering the video natively in the target ratio is the only approach that neither loses nor invents content.

What does 'AI reframe' actually mean?

In most tools, AI reframe means AI-guided cropping: a subject-tracking model follows the main subject and moves a crop window across the frame, permanently discarding everything outside it. Adobe Premiere's Auto Reframe has worked this way since 2019, and Opus Clip, CapCut, Klap, Vizard, Kapwing, Veed, and Descript use the same mechanism. A smaller generative category expands the frame instead (outpainting), and Riffkit's reframe re-renders the whole video in the new ratio with each shot recomposed for the new frame.

Crop vs outpaint vs re-render: the three ways to change a video's aspect ratio

Search “how to change video aspect ratio without cropping” and you'll find dozens of tools claiming to solve the same problem. They don't. Every tool that converts a video from one aspect ratio to another uses one of exactly three mechanisms: crop, outpaint, or re-render. Cropping discards pixels, outpainting invents pixels, and re-rendering recomposes the shot. Once you know which mechanism a tool uses, you can predict what your video will lose, or what it will gain that was never really there, before you press export.

The distinction matters most for ad creative. If your video carries information in its composition (a product on the table, captions near the edge, two people in frame), the mechanism you pick decides whether that information survives the conversion.

Three mechanisms, not thirty tools

The tool landscape looks crowded, but the underlying engineering is not. Strip away the marketing labels (“AI reframe,” “smart crop,” “magic expand”) and aspect ratio conversion reduces to three operations:

Crop: move a window across the original frame and discard whatever falls outside it.
Outpaint: keep the original frame intact and generate new content beyond its edges.
Re-render: use the original video as a reference and regenerate it natively in the target ratio, recomposing each shot.

Each mechanism makes a different bargain with your footage. Here is what each one actually does, which tools use it, and what it costs you.

Mechanism 1: Crop — track a subject, discard the rest

Smart cropping is a subject-tracking model driving a crop window. The model identifies the most important element in frame, usually a face, and slides a window of the target aspect ratio to keep that element centered as the video plays. Everything outside the window is permanently discarded.

This is the dominant mechanism by a wide margin. Adobe Premiere Pro's Auto Reframe has worked this way since 2019, built on Adobe Sensei. Opus Clip's ReframeAnything, CapCut, Klap, Vizard, Kapwing, Veed, and Descript all convert aspect ratios the same way, per their official documentation (2024–2026). When a tool advertises “AI reframe,” it almost always means AI-guided cropping.

The geometry is unforgiving. Converting a 9:16 vertical into a 1:1 square keeps only a 9:9 slice of the frame: you lose close to half the picture. The casualties are predictable. Captions burned near the top or bottom get cut. A product at the edge of the table disappears. The second person in a two-shot exits the frame entirely. Subject tracking keeps a subject in view; it cannot keep everything that mattered.

The non-AI fallback is padding (letterboxing): keep every pixel and fill the leftover space with bars or a blurred backdrop. Nothing is lost, but your video renders smaller inside the placement and visibly reads as footage made for a different format.

Use crop when nothing near the edges matters. For a centered talking head against a plain wall, it is fast, cheap, and genuinely good enough.

Mechanism 2: Outpaint — keep the pixels, invent the edges

Video outpainting inverts crop's bargain. The original frame stays untouched in the center, and a generative model paints new content outward until the target ratio is filled. Nothing you shot is discarded; the question becomes what gets added.

This mechanism defines the whole “generative aspect ratio” category. Runway's Expand Video, shipped on Gen-3 Alpha Turbo in November 2024, describes it as changing the aspect ratio “without changing the underlying content.” Luma's Dream Machine Reframe, released on Ray 2 in May 2025, promises that the “main subject stays perfectly intact” while the model “fills in around the edges.” Kling's Extend and Pika's Expand Canvas take the same approach, along with a long tail of smaller tools.

The tradeoff: everything beyond your original frame is hallucinated. The model invents a plausible continuation of the scene, and three failure modes recur:

Invented content. The fill can include things that were never there: an object on the counter, an extra hand, a background that contradicts the real location. For ad creative, footage showing things that don't exist is a real liability.
Style seams. Generated fill can drift from the original footage in grain, lighting, or sharpness, leaving a visible boundary where real ends and generated begins.
Unredesigned composition. The shot was composed for its original ratio. Outpainting gives you that same composition with wider margins, not a shot composed for the new format. A 9:16 vertical expanded to 1:1 is still, compositionally, a vertical video with generated wings.

Use outpainting when the edges are low-stakes. Scenery, ambience, b-roll: if a hallucinated stretch of grass or wall can't hurt you, keeping your original pixels intact is a fair deal.

Mechanism 3: Re-render — recompose every shot for the new frame

The third mechanism doesn't move a window or patch the borders. It treats the finished video as a spatial reference and regenerates the entire video natively in the target aspect ratio, shot by shot. Each segment of the source guides the generation: same scene, same performance, same pacing. What changes is the framing, because every shot is recomposed for the new frame instead of being adapted from the old one. The result is near frame-identical to the source, with only the framing rethought.

Nothing is discarded, because there is no crop window. Nothing is invented at the margins, because the frame is generated as one composition rather than extended at its edges. The constraint is equally clear: re-rendering is a full generation pass, so it only applies to video that can be regenerated, which in practice means AI-generated video where the pipeline still holds segment-level references. You can't re-render arbitrary phone footage.

This is the path Riffkit's reframe takes. When a 9:16 riff proves itself as a winning ad, you reframe it to 3:4, 1:1, or 4:5 for other placements. Each ratio is generated as its own video, recomposed shot by shot and billed by the second like any render, and ratios you've already generated are automatically skipped. Across the 24+ reframe and outpainting tools we surveyed for this piece, cropping and outpainting were the only mechanisms we could verify in official documentation; some tools don't document their approach, so we'll put it precisely: re-rendering is, at minimum, rare.

The three mechanisms side by side

	Crop	Outpaint	Re-render
Original pixels	Center kept, rest discarded	All kept, untouched	Used as reference, regenerated
New edges	None (edges are removed)	AI-generated fill	Composed natively in the new frame
Composition	Same shot, tighter window	Same shot, wider margins	Recomposed per shot
Content-loss risk	High: captions, products, second subjects	None	None
Invented-content risk	None	Real: hallucinated fill, style seams	Low: frame generated as one composition against the source
Works on	Any footage	Any footage	AI-generated video with segment references
Representative tools	Premiere Auto Reframe (2019), Opus Clip, CapCut, Klap, Vizard, Kapwing, Veed, Descript	Runway Expand Video (2024), Luma Reframe (2025), Kling Extend, Pika Expand Canvas	Riffkit reframe

How to choose

Each mechanism has an honest home.

Choose crop when the subject is centered and the composition is simple. A talking-head explainer with the speaker in the middle loses nothing meaningful at the edges. Crop is the cheapest and fastest mechanism, and for this footage it is genuinely enough.

Choose outpaint when the edges are low-risk background. Landscape b-roll, ambient mood shots, footage where the margins are texture rather than information. Keeping your real pixels and accepting generated scenery is a reasonable trade.

Choose re-render when the composition carries information. Ad creative usually does: a product held at the frame's edge, burned-in text, a price overlay, a demo where the hands and the screen both matter, a two-person skit. Crop will delete some of that information. Outpaint will keep it but may invent new information right next to your product. Re-rendering is the only mechanism that neither loses content nor invents content, which is why Riffkit's reframe is built on it.

One variable sits underneath all three choices: which ratios you actually need. Meta Feed's workhorse is 4:5; TikTok and Reels run on 9:16. The placement-by-placement breakdown is in our aspect ratio guide for TikTok, Reels, and Meta.

Convert the winner, not the guess

The deeper mistake isn't picking the wrong mechanism. It's converting a video before you know it deserves converting: every extra ratio is another asset to render, traffic, and manage. The workflow that pays is to make the 9:16 version first, run it, and let TikTok or Reels tell you whether it wins. Then spread the winner across placements. We laid that workflow out in one winning video, every ad placement.

That is the shape of Riffkit's pipeline: riff a proven video into your own 9:16 ad, validate it where vertical video lives, then reframe the winner to 4:5, 1:1, or 3:4, recomposed shot by shot rather than cropped or padded. You validate the formula once. The framing adapts to wherever the ad runs.