Video to shorts, reels and clips cutter

A tool that cuts a long video into short vertical clips. Speech is recognized in the browser, complete moments are picked from the transcript, and every chosen moment becomes its own file: the camera follows whoever is talking, captions are in the frame, the result is a finished 1080×1920 MP4.

How to cut a video into shorts

5 steps
1

Upload a long video: a podcast, lecture, stream or interview.

2

Check the spoken language and the clip length, and choose how to pick moments.

3

Wait for the transcript — it runs in the browser, and the model downloads once.

4

Look through the moments: uncheck the extra ones, fix the titles, move the edges by a phrase and preview the frame.

5

Choose the format, framing and caption style, then press "Build clips".

🔥 Trending right now
🗣️ Text to Speech ✨ Video Enhancer ✂️ Trim Audio ✨ Audio Enhancer 🎭 Voice Changer 🎙️ Vocal Remover 🎬 Create Video from Audio 🎬 Trim Video

Upload a podcast, lecture or stream and get finished vertical clips

Drop a video here or choose a file
MP4, MOV, WebM, MKV ·

Premium includes up to 35 a month; beyond that or without it — 350 tokens per video.

More options

MP4, MOV, WebM, MKV. A podcast, lecture, stream or interview — anything with speech.

The camera follows the speakerFaces are found in every moment and the lips show who is talking: the vertical frame cuts to them the way an editor would.
You see why a moment was pickedEvery moment carries its quote and the named signals, not an unverifiable percentage.
A batch at onceSeveral clips in one pass, saved to a folder or an archive instead of one by one.
Don't know what to try?
1030+ tools — open a surprise
🎲 Surprise me

Features

The camera follows whoever is talking: faces are found in each moment, the lips show the active speaker, and the vertical frame cuts between people the way an editor would — or stacks two people one above the other.
Moments are picked by the shape of speech — a thought that starts, a phrase that ends, density, pauses, a question with its answer — and each signal is named next to the moment; smart picking reads the meaning instead.
Every moment shows its quote and an editable title, and its start and end move phrase by phrase, so a clip never cuts a word in half.
Captions are burned in at 1080×1920 in nine styles, including karaoke fill and big uppercase, and sit above the app buttons of Shorts, Reels and TikTok.
Recognition, framing and building run on your device: FFmpeg through WebAssembly pulls the audio, WebCodecs rebuilds the frames, and several clips go out in one pass to a folder or an archive.
Ready SRT or VTT subtitles are accepted instead of recognition — when the transcript already exists, no model is needed.

What clips come out of

  • Podcasts and interviews: one answer with a complete thought
  • Lectures and webinars: one idea explained
  • Streams and reviews: a viewer question with the answer
  • Talks and presentations: a claim with numbers
  • Lessons and walkthroughs: one finished step
  • Live streams: a short answer to a frequent question

Formats and limits

MP4, MOV, WebM and MKV are accepted; the size limit adapts to the device memory. A clip runs from 15 to 90 seconds as found, and up to 3 minutes when you move its edges by hand. Finished clips are MP4 files at 1080×1920 (or 1080×1080 and 1080×1350) with H.264 video and AAC audio. Building needs WebCodecs, which current Chrome, Edge, Safari and Firefox have.

How this differs from cloud clippers

A cloud service first takes the whole file, and only then starts working: a two-hour podcast travels to it longer than it takes to cut. Here the order is reversed — the transcript and the clips are made on your device, and in "Smart picking" mode the text of the transcript goes to a model, not the video. The second difference is honest picking: instead of an unverifiable virality percentage, every moment carries its quote and the signals it was chosen by.

Frequently asked questions

Why are clips built in the browser instead of the cloud?

Because a two-hour podcast travels to a cloud longer than it takes to cut. FFmpeg through WebAssembly pulls the audio and WebCodecs rebuilds the frames — on an ordinary laptop that beats waiting for an upload to finish. In "Smart picking" mode the transcript text with its timings goes to the server, not the video; the recipients are listed on the AI providers page.

Why is there no virality percentage on the moments?

Because nobody has shown that such a score predicts views. The comparisons published so far were run by clipping services themselves, and in them high-scored clips did not reliably beat low-scored ones. A number you cannot check looks like knowledge and is not, so each moment here shows its quote and the reason it was picked: a complete phrase, a question with its answer, dense speech. That takes two seconds to verify with your own eyes.

Two people sit in a wide shot — who ends up in the vertical frame?

The one who is talking. In each moment the tool finds the faces, measures how much each mouth moves while speech is going on, and cuts the 9:16 frame to the active speaker; a shot lasts at least a second and a half, so the picture does not jump on every nod. If you want both on screen, choose "Two people stacked": each gets half of the frame at the same size. With no face in the moment — slides, a screen recording — the whole frame goes on a blurred background.

What happens to a two-hour video on a phone?

Recognition and building load the device, and a phone freezes a backgrounded tab: long runs there reach the end less often than on a computer. When a video is long for a phone, the tool says so before the work starts. The way around it is to bring ready SRT or VTT subtitles: recognition is then skipped entirely and only the chosen moments are built.

A clip starts in the wrong place — what now?

Moment boundaries come from the transcript: the start sits after a finished phrase and the end at the end of one. Next to every moment there are − and + buttons for the start and the end: each press moves the edge by a whole phrase, so the clip still starts and ends on a pause. If the speech was recognized poorly (a noisy recording, music, several people at once), choosing the language by hand or bringing your own subtitles helps.

Why do the captions differ from what was said?

Captions are drawn from the same transcript the moments were picked from, so recognition mistakes show up in the frame too. Russian speech is handled by GigaAM, other languages by whisper. When the frame needs verified text, bring your own subtitle file and exactly that text lands in the clip.

💡 Want us to improve this tool just for you?

We can! Just send us a quick message with your idea. If you'd like to discuss it in detail, leave your email and we'll get back to you. You can stay anonymous.

You May Also Like

📱 Reels Maker 🎥 Video Cropper 🎬 Trim Video 🎬 Subtitles & Transcript 🎬 Music Video Maker 🤬 Video Profanity Remover 🔤 Remove Text from Video 🔄 Video Flip & Mirror

How do you rate this tool?

Thank you for your rating!
Want to share more? Leave a comment!
Thank you! Your comment will appear after moderation.
Who is this tool for?
🎙️ Podcasters 🎬 Video Creators 📡 Streamers & Broadcasters ✨ Content Creators

By uploading content you confirm that you are entitled to disclose its contents to third parties, and you undertake not to include other people's personal data in it. See the Terms of Use for details. AI providers Tembrica is an independent product. We are not affiliated with, endorsed by or sponsored by the makers of the models and services listed here; their names are used only to say factually which engine runs a given tool.

Published Updated Author: