AI NewsOpen sourceAnnouncement
short-video-generator-AI turns any YouTube link into a vertical short and picks up 720 stars as a free OpusClip alternative
A new MIT-licensed Python project takes a YouTube link, transcribes it locally with faster-whisper, asks an LLM to rank the most viral moments, and renders them as vertical shorts with an optional hook at the start, and it has gained 720 stars in the four days since it appeared.
Image: GitHub
Why it mattersA working alternative to OpusClip and Vidyo.ai that runs on the creator's own machine, with three LLM providers to choose from, means the cost of a short becomes API tokens instead of a monthly seat, which shifts what small teams can afford to publish.
A four-day-old project on GitHub has picked up 720 stars for one specific reason: it is a working, MIT-licensed alternative to OpusClip and Vidyo.ai that runs on the creator's own machine. yukitorido released short-video-generator-AI on 23 September 2026 with a plain positioning line: "Designed for creators who want an alternative to short-video SaaS tools like OpusClip or Vidyo.ai for free."
What it does
You give it a YouTube link (any length) or a local video file. It downloads the source, transcribes the audio locally with faster-whisper, asks the LLM you have configured to classify the content type and pacing, then scans the transcript for hook moments, emotional peaks, opinion moments, revelations, conflict, quotables, story peaks and practical value. Each candidate gets a score from 0 to 100, overlaps are collapsed, the top N are selected, and FFmpeg auto-crops each one to a vertical 9:16 clip. There is an optional AI-generated hook line at the start of every clip, which you can turn off.
The choice of model, and the local part
The pipeline lets you pick the LLM at install time. The provider list is OpenAI, Google Gemini or MuAPI. Gemini has a free daily tier, OpenAI is paid, MuAPI is paid per call. Transcription is not a choice: the project uses faster-whisper locally on your own machine, in tiny, base, small, medium or large-v3 sizes, and switches to CUDA if the machine has a GPU.
The install is Python 3.10+ with a requirements.txt, one API key in a .env file for the chosen LLM, and FFmpeg on the system. The project also exposes an API you can call from your own scripts, and a local web UI.
Why the star count is real, not just a launch spike
The audience for short-form video tools is large and stated: creators paying a monthly fee to OpusClip or Vidyo.ai. Those SaaS tools do the same job (transcript to viral clip picks to 9:16 crop) and charge for it, so an open-source pipeline that produces the same output without a subscription is directly usable for the same people. 720 stars in four days on a repository whose README opens with a comparison to two named paid competitors is close to the strongest sign of demand a GitHub project can send.
The trade is that you provide the LLM key and run the pipeline yourself. For a creator who publishes shorts on a schedule and is paying a per-seat fee they only use a few days a month, the arithmetic is worth working out: OpenAI at gpt-4o-mini rates or Gemini on the free tier can process a long-form transcript for a small handful of cents.
For a small marketing team producing shorts from long-form video (podcast episodes, webinars, keynote clips, product demos) this is the kind of tool that changes what shipping short-form content actually costs. Once the API key is in place, the marginal cost of a batch of shorts is the LLM call, not another seat. The trade for that is running the pipeline yourself: faster-whisper, a Python environment and FFmpeg on the machine that produces the clips.
Source
Primary source: yukitorido/short-video-generator-AI on GitHub.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
