Most open source AI shorts generators fall into two families: faceless stock-footage pipelines and long-video clippers. Here are five on GitHub with the facts I checked today, what each one does and doesn't do, and where they fall down for code content.
Full disclosure first: I build ReelMint, a closed-source video studio for developers making vertical Shorts, so I have a stake here. This post is for someone who typed "open source ai shorts generator" into a search box and wants to know which repos are real, what they need to run, and whether any of them fit technical content. Stars, licences and last-commit dates below come from the GitHub API on 3 November 2026. Feature details come from each repo's README on the same day. I did not install or benchmark any of them, so I describe what they claim to do, not how well they do it.
The category splits into two families, and mixing them up is the most common mistake in search results.
What both families leave to you: editing a finished result, keeping a consistent brand look across videos, and judging whether the content is any good. Most also leave the publishing step half-built, and that last point matters more than it looks (see the YouTube policy section below).
All figures below were read from `https://api.github.com/repos/<owner>/<repo>` on 3 November 2026. Star counts move daily, so treat them as a snapshot.
MoneyPrinterTurbo (`harry0703/MoneyPrinterTurbo`): MIT, Python, about 137,900 stars, last push 2 Nov 2026. It takes a topic or keyword and generates the script, matches stock footage, adds voiceover, subtitles and background music, and renders an HD video. The README lists a web UI and an API, Docker support (`docker compose -f docker-compose.release.yml up`, UI on port 8501), and LLM options including OpenAI, Anthropic, Gemini and DeepSeek. Edge TTS needs no key; other voice providers need their own. The README says the minimum is 4 CPU cores and 4 GB RAM with no GPU. It lists publishing to TikTok, Instagram and YouTube Shorts as a feature but gives no setup details, so check that part yourself before relying on it.
OpenShorts (`mutonby/openshorts`): MIT, Python, about 7,400 stars, last push 2 Nov 2026. This is the long-video family: it detects moments, reframes with face tracking, adds subtitles and can dub. It also has an AI actor tool and a title and thumbnail tool. Self-hosting is `docker compose up --build`, then you enter keys in Settings. The README names Gemini as required for the AI features, fal.ai for the AI actor shorts, ElevenLabs for voice and dubbing, and Upload-Post for social posting (TikTok, Reels, YouTube Shorts). It ships an MCP server, at `localhost:8000/mcp` when self-hosted. There is also a hosted version; the README lists a free plan of 20 minutes a month with a watermark and paid plans from $12 a month.
ShortGPT (`RayVentura/ShortGPT`): MIT, Python, about 8,100 stars, last push 10 Feb 2025. The repo describes itself as an experimental framework for YouTube Shorts and TikTok channel automation, covering script, voiceover, footage, captions and rendering. I am not calling it abandoned. I am reporting that the last push was in February 2025, and that the README says documentation is still coming. The README does not list required keys, though it names OpenAI, ElevenLabs and Pexels as technologies used.
AI-Youtube-Shorts-Generator (`SamurAIGPT/AI-Youtube-Shorts-Generator`): MIT, Python, about 5,700 stars, last push 3 Nov 2026. It is a long-video clipper pitched as an open source alternative to Opus Clip. Give it a YouTube URL and it uses Whisper, an LLM for highlights, and automatic cropping to produce ranked 9:16 clips. API mode needs a `MUAPI_API_KEY`; local mode needs an OpenAI or Gemini key plus `ffmpeg`. The README does not mention uploading to YouTube. It also promotes a hosted per-clip API from the same ecosystem.
MoneyPrinterV2 (`FujiwaraChoki/MoneyPrinterV2`): AGPL-3.0, Python, about 32,300 stars, last push 15 Sep 2026. It bundles several automations, including a scheduled YouTube Shorts automator, a Twitter bot and affiliate tooling, configured through `config.json` and requiring Python 3.12. The AGPL licence matters if you plan to run a modified version as a service. I could not confirm the upload credentials it needs from the README.
I also looked for n8n and Make templates that wrap these steps. I did not verify any specific template well enough to list it, so I left them out rather than guess.
Developers asking about "youtube shorts automation" usually worry about getting a channel flagged. The relevant text is YouTube's channel monetization policies page, support.google.com/youtube/answer/1311392, which I read on 3 November 2026. A July 2025 update on that page renamed the policy from "repetitious content" to "inauthentic content".
What I took from it, in my own words. Automation and AI are not banned. The page says that if you use automated tools or templates, the final product still has to show your creative vision and give viewers educational or entertainment value. The things it names as a problem are content that looks made from a template, and AI content made with generic or unoriginal templates that gives the impression of mass production. Channels whose videos are interchangeable are described as not eligible to monetize. Note the scope: this is about eligibility for the YouTube Partner Program. Content that breaks the Community Guidelines is a separate matter and can be removed. I found no numeric threshold, such as a maximum number of uploads per day, and I won't invent one.
The practical reading for a pipeline: a script written by an LLM from a bare keyword, stock clips, a default voice and the same layout on every video is the pattern the page describes. A video built from your own repo, your own terminal output and your own explanation is not interchangeable with the next channel's. That is a judgement about the content, which no repo can make for you.
The topic-to-video family matches stock footage to sentences. Ask it for a video about a Postgres index and you get b-roll of servers and someone typing, not the query plan. There is no step where a real code snippet, a terminal session or a chart is rendered as the visual. The long-video family has the opposite problem: it needs a long recording to exist already, and it crops talking heads, not syntax-highlighted code that must stay legible at 9:16.
Three more gaps show up for developers specifically. First, editing: when scene four has the wrong variable name, you re-run the pipeline or open the MP4 in a separate editor. Second, consistency: nothing carries a code theme, font or colour palette from one video to the next. Third, accuracy: an LLM writing narration over stock footage has nothing checking that the code on screen is the code being described, because no code is on screen. Our own take on the editing problem is in editable AI Shorts generators, and the content side is in YouTube Shorts for developers.
ReelMint is not open source. It is a hosted product; you cannot read or self-host the code. What it is: a Remotion-based render pipeline where every generated scene is editable in a visual editor, with code cards, terminal cards and chart animations as scene types. It is driven by an MCP server that Claude Desktop, Claude Code and Cursor can connect to, with 130+ MCP tools. It is BYOK, meaning you bring your own OpenAI, Anthropic, ElevenLabs and similar keys, and there is a free tier. It publishes to YouTube Shorts. Instagram Reels publishing is pending Meta App Review, so I am not counting it.
On time, the only number I will stand behind: on the MCP path, a 60 to 90 second reel took me 45 to 60 minutes elapsed, of which 10 to 20 minutes were hands-on. That is one person, the person who built it, so a newcomer will be slower.
Yes. MoneyPrinterTurbo, OpenShorts, ShortGPT and AI-Youtube-Shorts-Generator are all MIT licensed per the GitHub API on 3 November 2026, and MoneyPrinterV2 is AGPL-3.0. Free to use does not mean free to run: most need at least one LLM key, and some need voice, video or posting keys, billed by those providers.
Partly. OpenShorts documents a REST API and an MCP server, and MoneyPrinterTurbo documents API docs at port 8080 in its Docker setup. Both can generate clips programmatically. Uploading is a separate step handled by YouTube's own APIs or a service like Upload-Post, which OpenShorts uses. I did not test any of these end to end, so verify the upload path before building on it.
YouTube's policy page does not ban automation or AI. It says the final video still has to show your creative vision and have educational or entertainment value, and it describes template-made, mass-produced, interchangeable content as ineligible for monetization. It reads as a monetization policy, not a ban on uploading. I found no stated upload-volume threshold.
If your content is technical, the easier route to non-interchangeable videos is to put real code and real output on screen. You can try that with the code explainer Shorts maker on the free tier.
Get one practical MCP-video tip when we publish the next post — no more than that.
Point your agent at a video pipeline that speaks MCP and make one small thing. Free founding-creator pilot — your keys, no watermarks.
Create your studio — free pilot