Long video to shorts,
decision by decision.

The hard part is a search problem.

The thing sitting on your drive is a recording nobody will watch end to end: a three-hour session, a stream VOD, a forty-minute run you kept because something good happened in it. Somewhere inside are maybe four to eight spans of twenty to fifty seconds worth posting. Everything else is walking, loading, menus, and the stretches where you were talking to nobody.

Three hours at 60fps is around 648,000 frames. You are hunting a handful of windows in that, and every route to a short starts by narrowing the hunt. Which is why "just cut it up" is bad advice: the cutting is the easy part, and the finding is where the hours go.

It also explains why tools in this category feel so unlike each other. Some help you find the moment. Some assume you already know it and help you frame it. A few try to do both. Before you pay for any of them, it is worth knowing which of the four decisions below you actually want handed off.

Four decisions between the file and a postable clip.

• 1. Which moments. The expensive one. A human does it by scrubbing and remembering. A tool does it by reading signals it can actually measure: the audio track getting loud or going quiet, what you said and how you said it, scene changes, and, where the tool was trained on the game, the state of the interface it learned to recognise. None of those signals are the moment itself. They correlate with it, which is why every automatic pick still needs a human glance before it goes out.
• 2. Where the clip starts and ends. Underrated and unforgiving. Two seconds too early and the viewer watches you reload while the hook has not landed. A second too late and the payoff is already gone. The end matters as much as the start: cut on the reaction, not on the walk away from it. Most disappointing automatic clips are not wrong about the moment, they are wrong about the boundary by a second and a half.
• 3. Reframing 16:9 into 9:16. Not a resize, a decision about what to throw away. Keep the full height of a 16:9 frame and crop it to 9:16 and you are left with about a third of the original width. The options are a centre crop, tracking a subject and panning the crop with it, a blurred pillarbox that keeps the whole frame small in the middle, or a split layout with facecam above gameplay. Gameplay is more forgiving here than a talking head, because the action tends to sit near the centre. Not always forgiving, though: a right-anchored health panel or a corner minimap can land outside the crop entirely.
• 4. Captions. Speech has to be transcribed, timed to the word, styled, and burned into the frame so it survives being reposted. For a podcast this is most of the value. For gameplay it is a smaller share, because a lot of a gaming short carries itself on what is happening rather than what is being said. If your footage is mostly silent, captions will not rescue a weak pick.

A fifth job hangs off the end of those four and rarely gets counted: writing a title and a description, tagging, and getting the file into a publishing queue. Small per clip, large per week, which is exactly the shape of work that quietly stops people from posting at all.

By hand, on someone's servers, or on your own PC.

Every product in this category is one of three shapes, and the trade-offs are structural rather than promotional. They do not go away with a better version.

Vendor prices checked 14 August 2026 from a European IP, so your currency and amount may differ. The $12 to $29 band is entry-tier headline pricing, and on some of those tools long-video clipping is not in the entry tier: Submagic sells it as a paid add-on on top of Starter, and StreamLadder's automatic clipping sits above Gold in the €25.41 ClipGPT tier. Editor cost depends entirely on which editor you already own.

The upload figure in row two is the one people underestimate. A three-hour 1080p60 recording commonly lands in the 15 to 30 GB range. On a 20 Mbit/s upstream connection, 20 GB is a shade over two hours of uploading before any processing begins. If your upstream is slower, or the vendor caps the file size, you end up transcoding the source down first, which costs you time and picture quality. That is arithmetic, not an accusation: for a ten-minute source it is irrelevant, and for anything under a gigabyte the server route is genuinely the fastest thing on this page.

Row three has the mirrored weakness. A local pass is bounded by the GPU you already own, which is why CutBench takes hours rather than minutes, and why it refuses to run at all on hardware it cannot use. None of that gets faster by paying more.

How to make shorts from long videos with no tool at all.

Worth knowing even if you end up automating it, because it is the reference every automatic result gets judged against. This is how to make shorts from long videos by hand, in the order that wastes the least time.

• Pass one: mark, do not cut. Play the recording at 2x or 4x with the audio on and drop a marker every time something makes you sit up. Do not open a single clip yet. Most people ruin the evening by editing the first good moment they find and never reaching the second half of the file. On a three-hour session this pass alone is realistically 40 to 60 minutes.
• Pass two: cull the markers. Go back over the list cold and delete half of it. A marker that needs you to explain the context is not a short. What survives should make sense to a stranger with no idea which game this is.
• Pass three: set the boundaries. For each survivor, find the earliest frame where the clip still makes sense and the last frame that still lands. Then trim two seconds off the front anyway and watch it again. This is where a clip is won or lost.
• Pass four: reframe and caption. Crop to 1080x1920, decide per clip whether a centre crop holds or the action drifts out of it, then transcribe and burn in the captions. An editor with automatic transcription makes this survivable. Without one, budget more time for captions than for everything else combined.
• Pass five: export and write the copy. Titles, descriptions, tags, and the upload itself. Small each time, and the reason a lot of finished clips never actually get posted.

Done properly, six shorts out of a three-hour session is an evening of work for someone who knows their editor. That number is the honest benchmark for everything else on this page: a tool worth paying for has to beat an evening, not beat perfection.

Pick by what your footage actually is.

This is where most pages of this kind quietly stop being useful, so here is the direct version, including the cases where the answer is not us.

• Podcasts, interviews, webinars, talking heads. Use a tool built for that, not this one. The signal in your footage is language: who says the interesting sentence, where the filler and the dead air sit, whether a face should be tracked inside the crop. On prices checked 14 August 2026, Opus Clip starts at $15 a month on Starter, billed monthly only; Vizard is $29 a month, or $14.50 a month billed yearly; Submagic starts at $12 per member a month billed yearly, with long-video clipping sold as an add-on rather than included in that entry plan. All three process on their own servers and run in a browser, so the machine on your desk and its operating system matter far less. CutBench is built for gameplay recordings and would be a poor fit for your footage.
• Stream VODs you never kept a local copy of. If the session only exists as a VOD on a platform, a local tool has no file to open, and Eklipse is built to work from a stream VOD library. One thing to know before you budget: checked 14 August 2026, it publishes no pricing page and the rate is shown at checkout. The longer comparison is on the Eklipse alternative page.
• You already know the moment and care how it is framed. A hands-on vertical editor beats any automatic verdict. StreamLadder is the obvious one, with Gold at €15 a month on the monthly toggle, checked 14 August 2026. If the decisions you want handed off are the third and fourth rather than the first, that is the tool shape you want.
• Gameplay recordings, on Windows, with an NVIDIA card. This is our case. You have a drive filling up with long sessions, you do not want them uploaded anywhere, you would rather pay once than monthly, and you can let the machine work for a few hours while you are not using it. That is what CutBench is for, and it is deliberately narrow.
• macOS, an AMD or Intel Arc GPU, or a console. Not us, and there is no workaround. CutBench will not install, and there is no processing path that skips the GPU. Any of the browser tools above will work instead; the PC clipping roundup lays the alternatives out with prices.

What CutBench does with a long recording.

CutBench takes the four decisions above and makes all four itself, on your own GPU, on footage that never leaves the machine. It does not record, it does not run while you play, and it is not a timeline you arrange clips in. You hand it a file and you get back a ranked list to approve or bin.

The models ship inside the build, so nothing downloads on first run and no network call is needed to process anything. There is no licence key either: ownership is a Steam client check. The free demo runs the identical pipeline and watermarks its exports, which is the only difference between it and the paid version.

Long video to shorts FAQ.

Related reading on the same job.

• How to clip a 3-hour stream. The same workflow written as a walkthrough of one long session, start to finish.
• What an automatic clip maker is actually reading. The signals behind decision one, and where they get it wrong.
• Best clipping software for PC. Seven tools in one table, with capture and cutting kept in separate rows.
• Offline clip maker. What changes when the recording never leaves your drive.
• Clip maker with no subscription. One-time pricing, and what it costs you in return.
• Clipping without an FPS drop. Why working after the session instead of during it costs zero frames.
• Rust clip maker. What a trained per-game profile changes about decision one.