Vertical Reframe: turn 16:9 footage into 9:16 clips for TikTok, Reels and Shorts

iReplay.tv Vertical Reframe converts landscape video into vertical clips without handing the framing decision to a black box. It reads the file, finds every scene cut, and shows you each shot with a 9:16 crop window already positioned on it. You move the window where it belongs, and it stays there for the whole shot. Free, no account, and every clip comes out at 1080×1920 H.264 with WebVTT subtitles.

Why the crop holds still instead of panning

A 9:16 crop of a 16:9 frame keeps only about 31% of the width, so every reframing tool has to answer the same question: which 31%. Most answer it by tracking motion and panning to follow it. On real footage that signal is far noisier than it sounds. Measured on a 12-minute news broadcast of 141 shots, the motion centroid wandered a median of 33% of the frame width within a single shot, because on a talking head it follows a moving mouth and then jumps to whoever moved last. A crop driven from that chases noise, and a pan that chases noise reads worse than no pan at all. So the default here is to hold: the window you place is the window that renders. The motion signal is still used, but for what it is actually good at, which is pre-positioning the window before you look at it. If a shot genuinely travels, you switch on "follow the action" for that shot alone.

Shots are not clips

Conflating the two is the usual mistake. Shots are the scene cuts, and each one gets its own framing, so a cut to a different composition does not inherit the last one. Clips are the vertical files you end up with, and their edges always land on a scene cut, never mid-shot, because a clip that opens half way through a camera move is the clearest sign of an automated job. You can make several clips from one source and queue the next while the last is still rendering. The expensive pass, reading the video, happens once, so the tenth clip costs no more of it than the first.

Shots that cannot be cropped

Some shots have no valid crop: a title spanning the frame, a wide establishing shot, a chart, a crowd, a sports field. Cropping those throws away more than it keeps, so the crop width is adjustable from a tight 9:16 slice all the way to the whole frame, and anything wider than the slice is letterboxed onto a blurred enlargement of the same picture rather than cut. The analysis proposes which shots need this; you can override it either way.

Two ways to decide the framing

Without an API key you frame each shot yourself and the crop holds. With your own key, a vision model picks the subject shot by shot and the crop follows it, and the watermark on free clips comes off. Anthropic, OpenAI and Google are supported along with around twenty OpenAI-compatible providers; the key runs your job and is deleted when it finishes. The rendering runs on our machines in both cases, so a key pays for the model rather than the compute.

Built for archives, not just for creators

Broadcasters, channel operators, event producers and VOD libraries hold thousands of hours of 16:9 that cannot go to TikTok, Instagram Reels, YouTube Shorts, Snapchat Spotlight or RedNote without reframing, and reframing by hand is a shot-by-shot job. Sixty seconds or less clears every one of those platforms at once. The clips also come out at exactly the top rung of the portrait ladder used by My TV Channel, so an archive can be turned into a 24/7 vertical channel with no upscaling and no re-authoring.

Checking a stream rather than re-cutting one? The Streaming Analyzer reads an HLS or DASH manifest against best practice, and the Video Quality Analyzer opens the frames themselves. To go the other way and create video from scratch, see the AI Ad Generator. This tool was built by Sylvain Corvaisier, a streaming and video engineer with 20 years in the industry. iReplay.tv is run by a collective of streaming and broadcast engineers; you can also hire streaming and broadcast specialists.