Upload a file or paste a link and get back short vertical clips built for Shorts, TikTok and Reels at once — reframed rather than squashed, captioned in the frame, cut on complete sentences, and each one graded so you are not guessing which to publish.
Real ClipSpeedAI output — every clip below was scored, cut, reframed to 9:16 and captioned automatically from the video above it.
Source video
15:27
The conversion is three decisions, and the tool makes all three.
Drag in an MP4 or MOV, or paste a link if the video already lives on YouTube. Webinar exports, Zoom recordings, screen captures, a camera file straight off a card — the input shape does not matter much, since everything is normalised before anything else happens. Paid plans take up to two hours per file, the free demo up to thirty minutes.
Conversion is not a resize job. Something has to choose which forty seconds of a fifty-minute recording deserve to be a short at all, and that is the decision being automated: the model reads the full transcript, marks the segments that hold up without setup, and cuts each one so it opens at the start of a sentence.
Each short comes back vertical by default, captioned and scored 0-100. Square and landscape versions can be rendered from the same clip when a destination wants a different shape, so a single conversion pass covers your Shorts slot, your TikTok slot and a 16:9 cut for a site embed without a second upload.
Four separate problems hide inside the phrase "convert to shorts". Here is how each is handled.
Going from 16:9 to 9:16 throws away about three quarters of the width. The lazy versions of this either letterbox the whole frame into a vertical box with grey bars top and bottom, or lock a crop to dead centre. Tracking the active speaker moves the crop where the content is, which is the only version that looks intentional.
A converted file is worthless if the forty seconds inside it were chosen at random. Segment selection is the part of conversion that actually determines whether anyone watches, and it is why this is a different product from a format converter that hands back the whole runtime in a taller box.
Every short-form app draws its own interface over your video: a caption and handle across the bottom, a column of buttons up the right. Text that sits in those zones is unreadable on the platform even though it looked fine in your export. Burned-in word-by-word captions are positioned to stay clear of that furniture.
Shorts, TikTok and Reels all want a vertical file and all differ in the details around it. Producing one properly framed vertical master with in-frame captions is what makes the same clip work in all three without a per-platform re-edit, which is where most of the manual time in cross-posting goes.
A talking head converts differently from a two-person call, and both convert differently from a screen recording where the slide is the point. Fill, fit, split, three-up, four-up, screen-share and gameplay picture-in-picture cover the realistic shapes of source material rather than pretending every video is a face in the middle.
Convert a long recording and you get more shorts than you have slots to post. Scoring each one 0-100 turns that surplus into a running order, and a running order is what stops a conversion job from becoming a folder you never open again.
Filler words, false starts and silence get removed in the same pass. Original footage is usually paced for a longer format, and short-form is unforgiving about the first three seconds, so cutting the slack is part of conversion rather than an optional extra.
The Brand Kit holds your fonts, colours and logo and applies them to everything converted, so twenty shorts from four different source recordings still look like they came from one account. Consistency is doing more work for recognition than any individual clip is.
The MCP connector means you can ask Claude to convert a video and get the shorts back in the chat. For pipelines that need conversion as a step — a media team turning every published recording into shorts automatically — there is an API behind the same engine.
A 16:9 call recording is the most common thing people need converted, and the worst thing to centre-crop. Speaker tracking and split layout are what make a recorded call survive the trip to vertical.
Years of landscape uploads convert into short-form supply without filming anything new. The YouTube Shorts maker covers that path specifically.
Long teaching recordings hold self-contained answers that work as standalone shorts. Converting them creates search-friendly entry points back to the full course.
All-hands recordings and training sessions get watched far more as ninety-second verticals on a phone than as an hour-long link in an email nobody opens.
A 21:9 capture loses even more width than 16:9 when it goes vertical. Picture-in-picture keeps the camera visible while the action fills the frame rather than being shrunk into a stripe.
Client footage arrives in every shape imaginable. A single conversion path that copes with all of it is worth more than a slightly better result on one specific format.
Type "convert video to shorts" into a search bar and half the results are format converters: pick an output size, press go, receive the same video in a different box. That solves a problem almost nobody has. Your fifty-minute recording is not too wide, it is too long, and no amount of resizing fixes that.
The decisions that actually make up a conversion are which segment to keep, where to point the frame within it, how to keep it legible with the sound off, and how tightly to pace it. All four have to be made for every clip. Doing them by hand takes fifteen to thirty minutes each, which is why most long video never gets converted at all rather than getting converted badly.
Handing those four decisions to a model changes the economics rather than the ceiling. A skilled editor will beat the automatic version on any single clip. The automatic version produces the twenty clips that would otherwise have been zero.
People assume each platform needs its own edit. Mostly they do not. A clean vertical clip at a standard high-resolution 9:16, with legible burned-in captions kept away from the edges, is accepted everywhere and looks right everywhere. The differences between the apps sit in metadata, length rules and posting behaviour rather than in the video itself.
Length is the one place to pay attention, and it is also the thing platforms change most often, so we do not print a table of limits that would be stale within a quarter. The durable advice is that shorter usually wins on all three, and that a clip built around one complete idea tends to land in the range every platform accepts anyway.
The other durable rule is safe zones. Each app overlays its own controls and text on your video, and anything you burn into the lower portion or the right-hand strip risks disappearing behind them. Captions are placed with that in mind, which is a small detail that quietly ruins a lot of otherwise fine exports.
It is not a general-purpose transcoder. If what you need is an H.265 file rewrapped as H.264 or a bitrate ladder for your own player, use ffmpeg or your existing encoder; nothing here is aimed at that job and you will be happier with a real tool.
It is also not for video with no spoken content. Selection reads the language of the recording, so a silent timelapse, a music video or a wordless montage offers nothing to choose from. The result would be arbitrary segments, and arbitrary segments are not shorts.
And it will not invent footage. There is no AI video generation, no synthetic stock footage and no machine-made narration — everything that comes out was in the source you supplied. That constraint is deliberate, and for anyone converting real recordings of real people it is the point rather than a shortcoming.
Burning text into the bottom third is the classic one. On a laptop preview it looks balanced and centred; on a real device it sits underneath a username, a description and a music ticker. Anything you want read has to live in the middle band of the frame, and that includes logos.
Padding a landscape clip into a vertical canvas is the second. Bars top and bottom are read by viewers as a signal that the clip came from somewhere else and was not worth reformatting, and they scroll before deciding whether the content is good.
A third one is converting the clip but not the pacing. The same forty seconds that works inside a fifty-minute video, where the audience has committed, dies in a feed where the first two seconds decide everything. Cutting the run-up is not optional in short form.
And the quietest failure is converting from an already-compressed re-upload. Every generation of re-encoding costs detail, and a short made from a screen recording of a stream of a compressed export looks exactly as rough as that chain implies. Go back to the highest-quality source you still have.
Do one conversion manually and you will know exactly what is being automated. Scrub the recording, mark the segment, cut it, drop it on a vertical canvas, keyframe the crop so the speaker stays inside it, generate captions, correct the four words the transcriber misheard, move the text out of the button column, tighten the pauses, export. Say you did that for a two-hour webinar: the cutting is perhaps twenty minutes per clip, and the watching that found those clips happened before any of it started.
Somebody doing that work will beat the automatic version on the individual clip almost every time. They will choose a braver in-point, keep the pause that lands, and notice that the useful half of an answer is the second half. None of that is in dispute and it is not what the comparison is about.
Most people who try to convert a back catalogue by hand produce four clips and then stop, and four clips is not a test of anything. Twenty decent ones tell you which two topics your audience reacts to. Four excellent ones tell you very little, because the variance in short-form performance is far wider than the quality gap between a hand cut and an automatic one.
So the division worth defending is to automate the pass, then spend your editing attention on the three clips that scored well rather than spreading it evenly over twenty. A two-second trim and a re-render costs almost nothing, and that is precisely where hand-editing earns its place.
If you already know your timestamps, none of this is worth paying for. A trimmer, ffmpeg, or whatever editor you already own will handle the shape change and the export, and the selection step you would be buying is one you have already performed yourself. For example, a webinar team that logs every question as it is asked has effectively pre-selected, and should be shopping for a cutter rather than a chooser.
If the source has no speech — a music video, a silent timelapse, a wordless product montage — selection has nothing to rank, because it reads language rather than pictures. In that case the honest alternative is a person with a timeline, and no amount of re-running changes the answer.
If what you need is genuinely a format conversion, a codec change, a rewrap or a bitrate ladder for your own player, then ffmpeg and HandBrake are free, better at it, and not competing with anything here. Pointing a content tool at an encoding job is the one misuse that reliably wastes an afternoon.
And if you produce one flagship recording a quarter rather than a weekly feed, an editor for hire will give you a better result than any pipeline. The economics only flip when the same job comes round every week: a quarterly showpiece is worth a person, an ongoing grind is worth automation. Worth knowing which of the two you actually have before you commit to either.
Start with your three best-performing long videos rather than your newest. Content that already proved it resonates is the safest supply, and if the converted shorts do not work from that material, they will not work from the rest either.
Convert in batches and review in one sitting. Switching between converting and posting all day is how the habit dies; an hour on Monday producing fifteen shorts, scheduled out across two weeks, survives a busy calendar in a way that daily improvisation does not.
Keep the clips that scored well but did not get posted. A converted library is an asset — when a week goes sideways, having eight finished verticals waiting is the difference between staying consistent and disappearing for ten days.
Re-convert selectively rather than exhaustively. If a short performs, the source it came from almost certainly holds two or three more, and going back to a proven recording beats working through the archive in chronological order.
Observations from operating the pipeline in production — not general advice.
A pasted link can be checked before anything is fetched, because the runtime is published metadata sitting next to the video. An upload cannot be: the browser streams the bytes first, and the duration only becomes readable once the whole file is on our disk. That ordering used to mean a fifty-nine-minute recording could finish a long upload over a hotel connection and only then be told the free demo stops at thirty minutes, which is the worst possible moment to learn it. The fix was a preflight call that sends the duration ahead of the file and gets a yes or a no before a single byte moves, and it is why the size of your source now matters far less than it did.
This is arithmetic rather than taste, and it settles more conversions than any option on the page. Take a 1920 by 1080 source: a full-height 9:16 window cut from it is 608 pixels wide, while a full-height 1:1 window is 1080. Vertical therefore throws away about 68 per cent of the original width and square throws away about 44. That gap is the reason a two-person conversation which cannot survive the vertical crop frequently survives the square one intact, and why trying 1:1 is worth a minute before you write off a wide two-shot. It also explains why a 21:9 gameplay capture is the hardest case here — the vertical window is barely a quarter of the frame it came from.
Everybody expects a re-upload to look softer, and it does. What surprises people is that the first thing to break is not the image, it is the choice of moment. Selection runs on a transcript, and every generation in the chain — a screen capture of a stream, exported again, uploaded again — takes another pass at the audio, which is the track nobody inspects before submitting. A slightly mangled transcript yields slightly wrong segment edges and a shortlist that is quietly worse in a way that reads as the model having poor judgement. In practice, going back one link in that chain to the best file you still hold fixes more converted clips than changing any setting here.
Those change the container shape and hand back the same content. You still have a fifty-minute video, now vertical, which no feed wants. They are useful when you already know your clip and only need a shape change; they do nothing about the part that takes the time.
Both can do everything described here and more, with total control, given the hours. The honest gap is repetition: converting one clip is a pleasant twenty minutes and converting twenty is a lost weekend. Use an editor for the piece that has to be perfect and automate the queue behind it.
Auto-reframe features handle the crop reasonably well and stop there. You still choose the segment, still add and time captions, still cut the filler, still export each one. Reframing was never the expensive step — deciding what to convert was.
The core loop is similar across the category, so judge on output. Two things here are uncommon: shorts can be produced from a live broadcast rather than a finished file, and the whole thing is callable from Claude or your own code. Convert one of your own videos and compare the framing before deciding anything.
Your back catalogue is the cheapest short-form supply you have. Take one old recording, run it through, and see how many usable verticals were sitting inside it.
⚡ Convert To Shorts — $1 trial
