OpenShorts GitHub (2026): Install and Deploy the Clipper

By Antigravity Skills editorial team · Last updated

Short answer: OpenShorts GitHub means mutonby/openshorts, an open-source AI tool that cuts a long video into vertical 9:16 clips with captions. It is MIT-licensed and has about 9,200 stars. To self-host it, clone the repo, run docker compose up --build and open localhost:5175. The repository also ships an agent skill.

Skill page: the agent skill has its own guide, OpenShorts agent skill: install and use, with the install commands and the flow it gives your agent.

Unofficial guide. Antigravity Skills is not affiliated with OpenShorts. The official sources are the repository github.com/mutonby/openshorts and the site openshorts.app. Facts below were checked on November 19, 2026 against the GitHub API, the README, the Docker files and the site. The project changes almost daily, so confirm commands at the source.

At a glance

What it isAI clip generator: long video in, 9:16 clips with captions and titles out
Repositorymutonby/openshorts
Stars and forks9,206 stars and 1,922 forks on November 19, 2026
CreatedDecember 19, 2025
Last pushNovember 18, 2026 (23:29 UTC)
LicenseMIT for the core app; the cloud/ folder has a separate commercial license
StackPython 3.11 and FastAPI backend, React 18 and Vite dashboard, FFmpeg, Docker Compose
Self-host cost$0 for the software; you bring your own API keys and hardware
Hosted versionopenshorts.app, the same software on the maintainers' GPU
Agent skillYes, in the repository at skills/openshorts/SKILL.md

Which OpenShorts GitHub repository is the right one?

Several GitHub repositories use the name. The README of mutonby/openshorts and the site openshorts.app both point at each other, so they are one project. These are the ones we saw in a GitHub search on November 19, 2026:

RepositoryWhat it isStars
mutonby/openshortsThe AI clip generator this page covers; homepage is openshorts.app9,206
mutonby/openshorts-pluginClaude and Cursor plugin by the same owner; connects to the hosted OpenShorts MCP server91
faris-sait/openshortsA separate project with the same name: it turns long YouTube videos or uploads into 9:16 clips431
jnMetaCode/openshortsA separate project, "OpenShorts (Kaipian)": topic in, finished short video out73
idhamsy/openshortlinkUnrelated: a URL shortener that runs on Cloudflare123

There are also many forks and copies of mutonby/openshorts under other accounts. We did not review them. If you want the project that ranks for the keyword, install from mutonby/openshorts.

Is openshorts.app the same project?

Yes. The site calls itself OpenShorts Cloud and the README describes it as "the same software on a GPU". The repository's homepage field is www.openshorts.app. The difference is who runs it:

Self-hosted (the repo)Hosted at openshorts.app
PriceFree, MITFree plan, paid plans from $12 a month, per the site
API keysYou bring Gemini, and optionally ElevenLabs, fal.ai and Upload-PostGemini is included
WatermarkNoneOn the free plan
Speed5 to 8 minutes for an 8-minute video on CPU, per the READMEAbout 50 seconds on the maintainers' GPU, per the README
Posting to socialsWith your own Upload-Post keyBuilt in on paid plans
AutopilotNot includedPaid plans

Plans and prices change. Read the current pricing on openshorts.app before you rely on the numbers above.

What OpenShorts does

The README describes three tools in one app, and the first is the one most people want:

  • Clip Generator. You paste a YouTube link or upload a file. It transcribes the video, finds 3 to 15 moments of 15 to 60 seconds, and renders them as vertical clips. Each clip gets word-level captions, a hook line, a virality score and titles. The README says a clip editor lets you cut by transcript word and re-render.
  • Smart reframing. It follows faces (MediaPipe and YOLOv8 per the README), blurs the background for wide shots, and stacks two speakers when two people are on camera.
  • AI Shorts. Marketing videos with AI actors, built from a product description or URL. This needs a fal.ai key and costs per video, which the README puts at about $0.65 in Low Cost mode.
  • YouTube Studio. Titles, thumbnails and descriptions with chapter timestamps.
  • Voice dubbing into 30 or more languages, with your own ElevenLabs key.

Requirements

From the README, the Dockerfile and the .env.example file:

  • Docker and Docker Compose. This is the documented way to run it. The Dockerfile installs FFmpeg and Node.js inside the image, so you do not install them yourself.
  • A Google Gemini API key. The README says it is required for all AI features and the Gemini free tier is enough to start. The exception is below.
  • No Google key, if you run a local model. Set LLM_BASE_URL to an OpenAI-compatible server such as Ollama, LM Studio or vLLM and the moment picker runs on your model. Layout detection and silent videos still need Gemini.
  • Optional keys. fal.ai for AI Shorts, ElevenLabs for dubbing, Upload-Post for posting to TikTok, Instagram and YouTube, and AWS S3 for clip backup.
  • Memory. The site says "realistically 8GB of RAM and a multi-core CPU".
  • GPU. Optional. The default image is CPU-only. An NVIDIA card with NVENC speeds it up, see below.

Install OpenShorts locally

These are the README's steps:

git clone https://github.com/mutonby/openshorts.git
cd openshorts

# optional: only needed for S3 backup
cp .env.example .env

docker compose up --build

Then open http://localhost:5175, go to Settings and paste your API keys. The README says keys are encrypted in your browser's local storage and not stored on the server.

The compose file starts three containers, per the file on November 19, 2026:

ServicePortRole
backend8000FastAPI server, also serves the /mcp endpoint and the OpenAPI docs at /docs
frontend5175The dashboard, which maps to Vite's port 5173 inside the container
renderer3100Remotion render service

Deploy OpenShorts on a server or GPU machine

This is where the documentation is thin, so we separate what is written from what is not.

Documented: GPU on your own machine. The README gives an NVIDIA path. Install the driver and the NVIDIA Container Toolkit, create a docker-compose.override.yml that builds the backend with GPU: "1" and reserves the GPU, and set these in .env:

WHISPER_MODEL=large-v3-turbo
WHISPER_DEVICE=cuda
WHISPER_COMPUTE=float16
FFMPEG_ENCODER=auto

The README says 8 GB of VRAM is enough for that model plus the detection models, and that a CUDA error in transcription retries once on CPU.

Documented: the same compose file on a Linux server. Because the supported route is docker compose up --build, you can run it on a VPS that has Docker. The README says only that Docker is required. It does not give a server guide.

Not documented in the README: a reverse proxy, HTTPS, a domain, running the dashboard as a production build, or a one-click deploy to a cloud host. The frontend service in docker-compose.yml builds Vite's dev target, so it is a development setup as shipped. If you expose it to the internet you need to add your own proxy and access control. We did not deploy it to a public server.

Other compose files exist (docker-compose.cloud.yml, docker-compose.worker.yml). The cloud file says it adds PostgreSQL and the settings for the hosted service, and the README says the cloud/ folder has a commercial license that forbids offering it to third parties as a paid hosted service. Self-hosting the core app does not need that folder, per the README.

Use OpenShorts with an AI coding agent

OpenShorts exposes the same pipeline to agents in three ways, all from the README:

  • MCP server. A self-hosted instance serves /mcp. You can add it to Claude Code with claude mcp add --transport http openshorts http://localhost:8000/mcp, or over stdio with claude mcp add openshorts -- python mcp_stdio.py. The tools include process_video, get_job_status, list_clips, add_subtitles, recut_clip and publish_clip.
  • CLI and REST API. pip install openshorts gives a command-line client that talks to the API. Point OPENSHORTS_API_URL at http://localhost:8000 for a self-hosted instance.
  • Agent skill. The repository includes skills/openshorts/SKILL.md, which tells an agent the submit, wait, list, edit and publish flow. It works in Claude Code and other skill-capable agents. Install commands and what the skill covers are on our OpenShorts agent skill guide. For general setup, see how to install skills in Antigravity, Claude Code and Codex.

Limits to know before you use it

  • Rights. The skill's rules say to submit only videos you have rights to, and the process_video tool requires a confirm_rights acknowledgement. Clipping someone else's channel is your risk, not the software's.
  • Speed on CPU. The README's own figure is 5 to 8 minutes for an 8-minute video, so a long podcast takes a while without a GPU.
  • Hosted features are not all in the repo. Autopilot and built-in social connections are paid hosted features, and the cloud/ code is source-available, not MIT.
  • Posting needs a third party. Self-hosted auto-posting goes through an Upload-Post account, which has its own limits and pricing.
  • Fast-moving project. The latest commits are from the day before we checked, and some change billing rules for the hosted app. Pin a commit if you need a stable setup.
  • We did not run it. This guide comes from the repository files and the site. We did not build the Docker image or process a video ourselves.

FAQ

What is the OpenShorts GitHub repository?

It is mutonby/openshorts, an open-source AI clip generator. It was created on December 19, 2025 and had 9,206 stars when we checked.

Is OpenShorts free?

The self-hosted software is free under the MIT license, and you pay only for the AI services you choose to use. The hosted version at openshorts.app has a free plan with a watermark and paid plans.

How do I deploy OpenShorts with Docker?

Clone the repository and run docker compose up --build, then open http://localhost:5175 and enter your Gemini key in Settings. The README does not describe a production server setup.

Does OpenShorts need a GPU?

No. The default image runs on CPU. An NVIDIA GPU makes it much faster, and the README gives the steps.

Does OpenShorts need a Gemini API key?

Yes for the default setup. If you point LLM_BASE_URL at a local OpenAI-compatible model, the moment picker no longer needs one, but layout detection and silent videos still do.

Is openshorts.app the same as the GitHub project?

Yes. openshorts.app is the hosted version of the same software, and the repository links to it as its homepage.

Does OpenShorts have an agent skill?

Yes. The repository contains skills/openshorts/SKILL.md, and the project is listed on skills.sh with the command npx skills add mutonby/openshorts.

Sources