Online, nothing to install
The OpenShorts app in your browser: paste a YouTube link or upload a file and get clips back. On the free plan it works as a free AI clip maker for 20 minutes of source video a month, with a watermark.
OpenShorts is an MIT-licensed open-source AI video clipper that turns long videos into 9:16 Shorts with word-level captions; it can be self-hosted or used online at openshorts.io.
Free account · 20 minutes a month · watermark on free exports · also available to self-host
Clips sized for every major short-video platform
OPENSHORTS.IO
OpenShorts is an MIT-licensed AI clipper that turns long video into vertical Shorts. People also look for it as OpenShorts AI, and it does the whole job of an AI clipping tool: it reads the transcript, picks the moments worth posting, reframes each one to 9:16 around the speaker and burns in word-by-word captions.
In
A YouTube link or a video file you have the right to use.
Out
Vertical 9:16 MP4 clips, each with a title and captions.
Code
Public and MIT-licensed, so the same pipeline runs on our servers or on yours.
Pick the one that matches how much you care about uploads leaving your machine.
The OpenShorts app in your browser: paste a YouTube link or upload a file and get clips back. On the free plan it works as a free AI clip maker for 20 minutes of source video a month, with a watermark.
Run OpenShorts self-hosted with Docker on your own machine. No watermark and no minute cap; you bring the hardware and a model key.
Drive it from an AI agent. Our hosted MCP server lets Claude, Cursor and other assistants clip videos with your API key; the open source build has its own REST API and CLI.
Not sure which route fits? The full online vs self-hosted comparison is in the self-host guide.
The whole stack is four commands and one model key.
$ git clone https://github.com/mutonby/openshorts.git$ cd openshorts$ cp .env.example .env$ docker compose up --build# then open http://localhost:5175Clone the repository and start the containers. Copying the example environment file is only needed for optional settings. The dashboard comes up on port 5175; paste your Gemini API key under Settings there.
Docker, 8 GB of RAM and a Gemini API key are enough to start; a GPU makes it faster but is optional. On a 4-core, 8 GB server with no GPU, a 3-minute vertical render took about 11 minutes in our test.
Install checks, Windows and WSL notes and common errors live in the self-hosted setup guide.
Three steps decide whether a clip is worth posting: where it starts, what stays in frame, and whether the captions keep up.
The transcript is scored window by window, so clips start where a sentence starts. The strongest windows are shortlisted and each one comes back with a start, an end and a title, which is the part an AI clip maker has to get right. Each clip also gets a score from 0 to 100 and a one-line reason, so you can see why it was picked and post the strongest first. An opening hook line is written for the first seconds, and ready-to-paste TikTok and Instagram captions come with every clip.
Every shot gets its own framing. A wide shot keeps the full frame; a talking head is tracked and held in the centre of the 9:16 crop, so the speaker stays in frame when the layout turns vertical. A cover frame is chosen from inside the clip, so the thumbnail is never a black first frame.
Every word carries its own timestamp, so the highlighted word is the one being spoken. Captions are burned into the export, and the timing comes from the same transcript the clips were picked from. Six caption styles are built in, from bold karaoke-style highlights to a single word at a time or clean sentence-case subtitles, and you pick one before the clips are made. Burned-in captions mean the clip reads the same on every platform, with the sound on or off.
The export is a standard vertical MP4 (H.264, 9:16), the format YouTube Shorts, TikTok and Instagram Reels all accept, so one file covers the three.
06 · Licence
The repository is public. The licence is MIT, which allows commercial use. Clips exported from a self-hosted build carry no watermark.
Short answers to what people ask before they try it.
The project is written OpenShorts, one word. You may have seen it written as open shorts, openshots or openshort; it is the same open source clipper.
OpenShorts.io is an independent deployment of the same MIT repository; it is not operated by openshorts.app.
Self-hosting is free under the MIT licence and its exports carry no watermark. Online, the free plan covers 20 minutes of source video a month, exports at 720p and adds a watermark; paid plans remove it.
Docker, a machine with 8 GB of RAM and a Gemini API key. A GPU is optional: on a 4-core, 8 GB server without one, a 3-minute vertical render took about 11 minutes in our test.
Yes. The repository includes an MCP server, a REST API and a CLI, so an agent can submit a video and collect the finished clips. The self-hosted version is not metered.
Yes. Clipping online needs a free account; the free plan does not ask for a card.
More on who runs this site: About OpenShorts.io