OpenShorts · Released, not on our provider

OpenShorts GitHub: what the repository is, how to install it and what it costs (2026)

OpenShorts GitHub is the repository github.com/mutonby/openshorts, an MIT-licensed open-source tool that turns long videos into 9:16 clips, picking the moments with Google Gemini or a local model, which you start with docker compose up after adding your own API keys and which its README says costs only your hardware and the API usage.

Released, not on our providerLast updated:

Use it now

Chat

This chat starts with GPT-6 Luna, not OpenShorts.

≈ 1 credit per reply · See all models

New accounts get 5 free credits once, valid for 7 days. They work on the lower-priced models here; the priciest ones need credits you bought.

Start chatting — 5 free credits
Maker
OpenShorts
Type
Text
Released
No date given
OpenRouter id
None

Checked

OpenShorts: facts, with sources

  • The README of github.com/mutonby/openshorts calls OpenShorts an open-source AI video platform with three tools in one: a Clip Generator that turns long videos into 9:16 Shorts, TikToks and Reels, AI Shorts (marketing videos with AI actors) and a YouTube Studio for titles, thumbnails and descriptions. It also describes itself as the open-source alternative to OpusClip, a closed hosted clip service.Source: raw.githubusercontent.com · checked 2026-11-19
  • This page covers one repository: mutonby/openshorts on GitHub, whose homepage field points to openshorts.app. GitHub’s API shows it is owned by a personal account (mutonby), was created on December 19, 2025, was last pushed on November 18, 2026, and is marked as a fork of kamilstanuch/Autocrop-vertical, a separate YOLOv8 and FFmpeg tool for turning horizontal video vertical. On November 19, 2026 it showed 9,217 stars and 1,924 forks; those numbers change daily.Source: api.github.com · checked 2026-11-19
  • The repository has no numbered release: GitHub’s releases list for it was empty on November 19, 2026, and its tags list was empty too. There is no version number to pin, so a clone gives you the main branch as it is on the day you copy it.Source: api.github.com · checked 2026-11-19
  • On who maintains it: the openshorts.app pages carry the byline "By the OpenShorts team" and name no individual, and the README presents OpenShorts Cloud (openshorts.app) as the hosted version of the same software. The README does not list contributors or say who the maintainers are, so this page does not name anyone.Source: openshorts.app · checked 2026-11-19
  • License: the LICENSE file in the repository root is the MIT License, "Copyright (c) 2024 OpenShorts". It lets anyone use, copy, modify, merge, publish, distribute, sublicense and sell the software as long as the copyright and permission notice stay with it, and it gives no warranty.Source: raw.githubusercontent.com · checked 2026-11-19
  • One exception to the MIT license: the cloud/ directory (billing, managed keys and the hosted-service infrastructure behind the optional BILLING_ENABLED flag) is under a separate OpenShorts Commercial License. It lets you read, modify and self-host that code for personal or internal business use, and forbids offering it to third parties as a hosted, managed or paid service, including one that competes with openshorts.app. The README says self-hosting the core app never requires that directory.Source: raw.githubusercontent.com · checked 2026-11-19
  • By the README, the clip pipeline runs in these steps: take a YouTube link (fetched with yt-dlp) or a local upload; transcribe with faster-whisper (or Parakeet on a GPU) with word-level timestamps; find scene boundaries with PySceneDetect; have Google Gemini pick 3 to 15 moments of 15 to 60 seconds, each with a virality score; cut with FFmpeg; reframe to 9:16 with MediaPipe and YOLOv8 face tracking; burn in captions, a hook line and optional effects; then download or post. It names Gemini 3.1 Flash-Lite as the model that ranks the transcript.Source: raw.githubusercontent.com · checked 2026-11-19
  • What it needs to run, by the README: Docker and Docker Compose as the base, plus a Google Gemini API key ("required for all AI features") unless you point the moment picker at a local model. A fal.ai key is needed only for AI Shorts, an ElevenLabs key only for voiceover and dubbing, and an Upload-Post key only for direct social posting. Its own comparison table lists the self-hosted setup as "Docker, 8GB+ RAM, model downloads", and the Tech Stack section names Python 3.11 for the backend.Source: raw.githubusercontent.com · checked 2026-11-19
  • By the maker’s own figures (not tested by us), an 8-minute video takes 5 to 8 minutes on a typical CPU when self-hosted, about a minute with an NVIDIA card (the README’s example is an RTX 4060, any card with NVENC), and about 50 seconds on the maker’s own hosted GPU. The default Docker image is CPU-only.Source: raw.githubusercontent.com · checked 2026-11-19
  • The README says the only cloud call in the clip pipeline is the moment picker, and that it sends the transcript, never the video, to Gemini. To avoid a Google key, set LLM_BASE_URL (and LLM_MODEL) to any OpenAI-compatible server: Ollama, LM Studio, vLLM, llama.cpp server, LocalAI or OpenRouter all work, it says. Ollama needs a larger context than its 4,096 default, and the README says 7-8B models return valid JSON reliably while 3B ones do not.Source: raw.githubusercontent.com · checked 2026-11-19
  • Even with a local model, some steps still need Gemini, the README says: the automatic layout picker, the on-screen content detector, and silent videos that have no speech to clip by. Without a Gemini key those fall back to the plain face-tracking crop, and a silent video fails with a message that says so. Adding a key next to LLM_BASE_URL gives you both.Source: raw.githubusercontent.com · checked 2026-11-19
  • The README says the whole pipeline can be driven by agents and scripts: a built-in MCP server at /mcp (self-hosted at http://localhost:8000/mcp, or over stdio with python mcp_stdio.py), a REST API with interactive docs at /docs on any instance, completion webhooks, a zero-dependency CLI (pip install openshorts) and an agent skill file. A self-hosted MCP endpoint answers only while your machine is on.Source: raw.githubusercontent.com · checked 2026-11-19
  • On source videos and copyright, the openshorts.app FAQ says to clip only a video you own or have the rights or permission to use. It says the app asks you to confirm that before processing, that you are responsible for what you submit, that fair use depends on the country and is never guaranteed, and that re-uploading someone else’s video without permission is copyright infringement that platforms remove and can strike the account for. The README does not discuss it beyond crediting its own demo video’s CC BY license.Source: openshorts.app · checked 2026-11-19
  • The README’s Security & Performance list says the containers run as a dedicated non-root user, API keys are encrypted client-side and never stored server-side, old jobs are purged automatically (24 hours by default when self-hosted) and uploads are capped at 2GB. It also describes a job queue with a configurable concurrency limit (MAX_CONCURRENT_JOBS, default 5). We have not audited any of this.Source: raw.githubusercontent.com · checked 2026-11-19
  • OpenShorts is software you run, not an AI model, and it is not on OpenRouter’s model list (checked November 19, 2026), so this site cannot run it. This site is a text chat with other models and does not edit video: it cannot cut, reframe or caption a clip. It can help with the text around the workflow, such as titles and hook lines.Source: openrouter.ai · checked 2026-11-19 · second-hand

Our table

OpenShorts GitHub requirements and costs, part by part

  • The README’s "How Much Does It Cost?" section says self-hosting is free, you provide the machine, and you pay only for the AI APIs you use. Its figures: Gemini has a free trial and costs under $0.01 per 10-minute video; a local model costs $0 beyond your hardware; fal.ai is pay-per-use at about $0.50 to $1.50 per AI Short; ElevenLabs has a free tier and is pay-per-use; Upload-Post gives 10 free uploads a month; optional AWS S3 backup is about $0.023 per GB. The AI Shorts section gives two modes: Low Cost at about $0.65 a video and Premium at about $2. These are the maker’s own figures, not tested by us.Source: raw.githubusercontent.com · checked 2026-11-19
  • The hosted edition, OpenShorts Cloud, is priced on the pricing page dated October 30, 2026: a free plan (the first video up to 60 minutes, then 20 minutes a month, watermarked, no card), Starter at $12 a month for 100 minutes, Creator at $29 for 300 and Pro at $59 for 750, all without watermark, plus extra minutes sold as 60 for $9 or 200 for $25. Self-hosted has no minute meter. Prices on a page like this change; check it before you rely on it.Source: openshorts.app · checked 2026-11-19

Parts and costs

What the OpenShorts repository needs to run, which AI service it calls, and what each costs

People who search for the OpenShorts GitHub page mostly want to know what they have to install and what it will cost. The table follows the README line by line. The clip pipeline itself runs on your machine; the one cloud call is the step that picks moments, which sends the transcript, not the video, to Gemini unless you point it at a local model. This site cannot run OpenShorts and does not edit video, but it can help with the text parts: pasting a transcript into the chat and asking for candidate moments or hook lines takes about 1 credit a reply with GPT-6 Luna, Claude Haiku 5.5 or DeepSeek V4.1 Flash.

Our table: the repository README, Dockerfile and .env.example and the maker’s own pages, read on November 19, 2026; costs are the maker’s own figures, not tested by us · checked 2026-11-19
PartWhat it does in the READMECost or condition as printed
Docker and Docker ComposeRequired. The install path is git clone, then docker compose up --build; the backend image is built on Python 3.11 and is CPU-only by defaultNo license fee for the core app (MIT). The README’s self-hosted column lists the setup as "Docker, 8GB+ RAM, model downloads"
Google Gemini API keyPicks the 3 to 15 moments (15 to 60 seconds each) from the transcript; the README names Gemini 3.1 Flash-Lite. Also used for the layout picker, scripts and web research. Required for all AI features unless LLM_BASE_URL is setFree trial with generous limits; under $0.01 per 10-minute video (README figures)
Local model (LLM_BASE_URL)Replaces Gemini for moment picking: an OpenAI-compatible server such as Ollama, LM Studio, vLLM, llama.cpp server, LocalAI or OpenRouter. Only the transcript is sent. The layout picker, on-screen content detector and silent videos still need Gemini$0 on your own hardware. Ollama needs OLLAMA_CONTEXT_LENGTH=16384 (its default of 4096 truncates silently); 7-8B models return valid JSON reliably, 3B ones do not
faster-whisper and ParakeetLocal transcription with word-level timestamps. Parakeet (GPU image only) is optional, about 2x faster than Whisper on a GPU, covers 25 European languages and falls back to WhisperRuns in the container, no API fee stated. The commented example settings in .env.example are the "small" model, cpu and int8
YOLOv8 and MediaPipeFace tracking for the 9:16 crop: MediaPipe first, YOLOv8 as the fallbackRuns locally; the Dockerfile downloads the yolov8n weights while building the image
NVIDIA GPU (optional)Faster transcription and NVENC video encoding; needs the NVIDIA driver, the NVIDIA Container Toolkit and a docker-compose.override.ymlAbout 2 GB more in the image; 8 GB of VRAM is enough for large-v3-turbo in fp16 plus the detection models
fal.ai keyOnly for AI Shorts: actors with Flux 2 Pro, talking head with Hailuo 2.3 Fast plus VEED Lipsync (Kling Avatar v2 in Premium)Pay-per-use, about $0.50 to $1.50 per AI Short; Low Cost mode about $0.65 and Premium about $2 per video
ElevenLabs keyVoiceover for AI Shorts and dubbing into 30+ languages with voice cloning; you bring your own key, on the hosted Cloud tooFree tier available, then pay-per-use, billed by ElevenLabs
Upload-Post keyOnly for posting clips straight to TikTok, Instagram, YouTube and more; not needed if you download the clips10 free uploads a month to all networks, no credit card, then pay-per-use
AWS S3 (optional)Clip backup and the public gallery; set through the AWS variables in .envAbout $0.023 per GB
OpenShorts Cloud (hosted, optional)The same software on the maker’s own GPU, Gemini key included, with auto-posting and Autopilot on paid plansFree plan (first video up to 60 minutes, then 20 minutes a month, watermark); Starter $12 a month for 100 minutes, Creator $29 for 300, Pro $59 for 750 (pricing page dated October 30, 2026)

Source: raw.githubusercontent.comSource: raw.githubusercontent.comSource: raw.githubusercontent.comSource: openshorts.app

Our table

OpenShorts GitHub install steps, from the README

  • The shipped docker-compose.yml defines three services: backend (port 8000), frontend (host port 5175 mapped to the dashboard’s dev server, built with target "dev", with the source folder mounted into the container) and renderer (port 3100). The README gives this one install path and no separate production or reverse-proxy guide; its .env.example mentions Traefik and PostgreSQL backups only for the maker’s hosted service. We did not run it.Source: raw.githubusercontent.com · checked 2026-11-19
  • GPU acceleration is optional and NVIDIA-only. The README says to install the NVIDIA driver (nvidia-smi must work) and the NVIDIA Container Toolkit, add a docker-compose.override.yml that builds the backend with the build argument GPU set to "1" (about 2 GB more in the image), and set the Whisper model, device and encoder in .env. It says 8 GB of VRAM is enough for the large-v3-turbo model in fp16 plus the detection models, and that a CUDA error in Whisper retries once on CPU.Source: raw.githubusercontent.com · checked 2026-11-19

Install steps

How to install and deploy OpenShorts with Docker, step by step

This is the short version of the README’s Getting Started section; the commands are the README’s, the notes are ours. Keep your own keys out of the repository folder and out of screenshots. The README says the dashboard keeps keys encrypted in your browser, not on the server.

Our table: steps paraphrased from the repository README and docker-compose.yml, read on November 19, 2026; we did not run the install · checked 2026-11-19
StepWhat to run or doWhat the repository says about it
1. Clonegit clone https://github.com/mutonby/openshorts.git, then cd openshortsNeeds Git and Docker. There are no releases or tags, so you get the main branch as it is that day
2. Configure (optional)cp .env.example .envThe README says to edit .env with AWS keys if you want S3 backup. The example file also carries commented settings for Whisper, the video encoder, a local model, rate limits and retention. The provider keys are entered in the dashboard, not here
3. Launchdocker compose up --buildBuilds three containers: backend (port 8000), frontend (host port 5175, the dashboard’s dev server) and renderer (port 3100). Model downloads are part of the setup, so the first start takes longer
4. Open the dashboardGo to http://localhost:5175, open Settings and enter your API keysThe README lists Gemini, fal.ai, ElevenLabs and Upload-Post. Keys are stored encrypted in the browser’s localStorage. If you set LLM_BASE_URL, the dashboard stops asking for a Gemini key
5. Make clipsIn the Clip Generator, paste a YouTube link or upload a long videoEach clip comes with a hook line, a virality score and titles for each platform, and opens in a clip editor with the full transcript. Use only video you have the rights to
6. GPU (optional)Install the NVIDIA driver and Container Toolkit, add docker-compose.override.yml with the build argument GPU "1", set WHISPER_MODEL=large-v3-turbo, WHISPER_DEVICE=cuda, WHISPER_COMPUTE=float16 and FFMPEG_ENCODER=auto in .env, then docker compose up --build -dThe README gives commands to verify the GPU inside the container. FFMPEG_ENCODER=auto probes the NVENC encoder at startup and falls back to x264
7. Drive it from an agent (optional)claude mcp add --transport http openshorts http://localhost:8000/mcp for the self-hosted MCP server; pip install openshorts for the CLI, with OPENSHORTS_API_URL=http://localhost:8000Self-hosted needs no OpenShorts API key. The API has interactive docs at /docs on any instance. The endpoint works only while your machine is on

Source: raw.githubusercontent.comSource: api.github.comSource: raw.githubusercontent.comSource: raw.githubusercontent.comSource: openshorts.app

Our table

OpenShorts GitHub: repositories with similar names

  • The name is shared. A GitHub search for "openshorts" in repository names on November 19, 2026 listed, besides mutonby/openshorts, a Claude plugin by the same account, several forks, mirrors and copies of its README, and a different project that also calls itself OpenShorts (Kaipian), which builds short videos from a topic rather than clipping long ones. This page covers only mutonby/openshorts; the table below says what each of the others describes itself as.Source: api.github.com · checked 2026-11-19 · second-hand

Similar names

Which GitHub repository is which when you search for OpenShorts

If you land on a different repository from the one this page covers, check the owner name in the address. The table gives what each page says about itself and, where it can be seen, how it relates to mutonby/openshorts. It does not rate their quality or safety, and the list is not complete: a search lists more copies and mirrors.

Our table: each repository’s own GitHub description and README opening as read on November 19, 2026; we did not compare the code · checked 2026-11-19
RepositoryWhat its GitHub page saysRelation to mutonby/openshorts
mutonby/openshortsOpen-source AI clip generator, MIT, Docker; homepage openshorts.app; the one this page coversThe repository named in the README’s clone command
kamilstanuch/Autocrop-verticalA smart video converter using YOLOv8 and FFmpeg to convert horizontal video to vertical 9:16 formatGitHub marks mutonby/openshorts as a fork of it; the README does not mention this
mutonby/openshorts-pluginA plugin for Claude that turns long videos into vertical clips and publishes them through the hosted OpenShorts MCP serverSame owner account; it uses the hosted service, not your own Docker install
faris-sait/openshortsAn AI-powered vertical video generator that turns long YouTube videos or uploads into 9:16 short clips; MITIts README opens with the same wording and badges as an earlier mutonby README, and its badges link to mutonby/openshorts. GitHub shows no fork link for it
gameplayxx111/openshortsDescribes itself as a free and open-source AI video platform with Clip Generator, AI Shorts and YouTube Studio, self-hosted, no watermarksSame three tools as the early mutonby README; we only read its description
legendryblade/openshortsA self-hosted shorts pipeline with karaoke captions, Groq cloud speech recognition and language model, and a self-healing watchdogIts description says "Fork of mutonby/openshorts"; GitHub’s license detector does not name a standard license for it
jnMetaCode/openshorts (OpenShorts · Kaipian)A local-first, open-source short-video pipeline that starts from a topic: it writes the script, finds footage, adds a voice and captions and renders the video; Chinese README with an English version; MITA different project that shares the name; it does not clip long videos

Source: github.comSource: api.github.comSource: api.github.comSource: github.comSource: github.comSource: api.github.comSource: github.comSource: github.com

Nearest available

Alternatives to OpenShorts you can use now

$0.10 / $0.50 (per million tokens)

  • OpenAI: "Our most efficient model for focused, high-volume tasks".Source: developers.openai.com · checked 2026-11-11
  • Context 1,050,000 tokens; maximum output 128,000 tokens; text and image input, text output; knowledge cutoff May 18, 2026.Source: developers.openai.com · checked 2026-11-11
  • Reasoning effort can be set to none, low, medium (the default), high, xhigh or max.Source: developers.openai.com · checked 2026-11-11

$0.10 / $0.50 (per million tokens)

  • Anthropic’s model page says Claude Haiku 5.5 was released on October 7, 2026, is the latest Haiku, and will not be retired sooner than October 7, 2027. Anthropic’s launch post is dated the same day.Source: platform.claude.com · checked 2026-11-17
  • In its launch post Anthropic calls Haiku 5.5 "the cheapest, fastest, and most capable small model we’ve ever released", for high-volume, cost-sensitive tasks such as summaries, compactions, database queries and classification, and as a coding subagent alongside Opus 5.5 and Sonnet 5.5. It calls Haiku 5.5 its first Haiku-class model with an adjustable effort setting.Source: anthropic.com · checked 2026-11-17
  • Anthropic’s model page says it is built for high-volume, latency-sensitive work such as classification, routing, extraction and subagent tasks, and rates its latency "Fastest" in the current lineup (Fable 5.2 and Opus 5.5 are "Moderate", Sonnet 5.5 is "Fast").Source: platform.claude.com · checked 2026-11-17

$0.30 / $1.2 (per million tokens)

  • DeepSeek released V4.1-Flash on September 10, 2026.Source: api-docs.deepseek.com · checked 2026-11-11
  • Only the Flash version of V4.1 exists: DeepSeek has published no V4.1 Pro, and the Pro model on its price page is still deepseek-v4-pro (V4-Pro-0813).Source: api-docs.deepseek.com · checked 2026-11-11
  • DeepSeek’s pricing page says the API model deepseek-flash is now served by the V4.1-Flash-1017 snapshot.Source: api-docs.deepseek.com · checked 2026-11-11

FAQ

OpenShorts FAQ

What is the OpenShorts GitHub repository?

It is mutonby/openshorts, an open-source AI video platform whose main tool, the Clip Generator, turns long videos such as podcasts and streams into captioned 9:16 clips for Shorts, TikTok and Reels. It also has AI Shorts (marketing videos with AI actors) and a YouTube Studio. The name is shared by other repositories, including a different project called OpenShorts (Kaipian) that makes videos from a topic, so check the owner name in the address.

Who maintains OpenShorts and what is the license?

The openshorts.app pages are signed "the OpenShorts team" and the repository sits under the personal GitHub account mutonby; the README names no individual. The core app is under the MIT License ("Copyright (c) 2024 OpenShorts"). The cloud/ directory (billing and the hosted service) has a separate commercial license that allows personal or internal self-hosting but not offering it to others as a paid service. It has no numbered releases, so you build the main branch.

What do I need to run OpenShorts?

Docker and Docker Compose, a machine with 8 GB of RAM or more by the README’s own table, and a Google Gemini API key, or a local OpenAI-compatible model set with LLM_BASE_URL. Keys for fal.ai (AI Shorts), ElevenLabs (voiceover and dubbing) and Upload-Post (posting) are needed only for those features. An NVIDIA GPU is optional. Even with a local model, the layout picker and silent videos still need Gemini.

How do I install and deploy it?

The README has one path: git clone https://github.com/mutonby/openshorts.git, optionally cp .env.example .env, then docker compose up --build, and open http://localhost:5175 to enter your keys under Settings. The compose file runs the dashboard’s development server and the README has no separate production guide, so putting it on a public server is up to you; we did not run it.

How much does it cost to run?

Self-hosting costs your machine plus the APIs you use. By the README’s own figures, Gemini is under $0.01 per 10-minute video, a local model is $0, and each AI Short is about $0.50 to $1.50 on fal.ai. The maker’s hosted OpenShorts Cloud has a watermarked free plan and paid plans from $12 a month for 100 minutes (pricing page dated October 30, 2026). We have not tested these figures.

What are the risks and limits?

API bills grow with use, especially AI Shorts. Only clip video you own or have permission to use: the maker’s FAQ says re-uploading someone else’s video without permission is infringement and platforms can strike the account. Self-hosting is slower on a CPU (5 to 8 minutes for an 8-minute video, by the maker’s figures), and there are no version tags to pin. We have not audited the code or run it.

Can I use this site for the text parts of that workflow?

This site cannot run OpenShorts and does not edit video. You can paste a transcript into the chat, in pieces because messages have a length limit, and ask for candidate moments, titles or hook lines. On this site a typical reply to GPT-6 Luna costs about 1 credit (a credit is about $0.01 in the $5 pack), billed by its real token cost, at least 1 credit; Claude Haiku 5.5 and DeepSeek V4.1 Flash are about 1 credit each. We give no API key for the repository.

Is this the official OpenShorts website?

No. This page is unofficial and not affiliated with or endorsed by OpenShorts. Facts link to the pages we read.