AI NewsOpen sourceAnnouncement
short-video-maker turns a long YouTube video into vertical shorts, free
An open-source tool that pastes a YouTube link, finds the viral moments, cuts and captions them into 9:16 shorts, picked up 624 stars in ten days.
Image: GitHub
Why it mattersA marketer or founder who used to pay a short-video SaaS by the month can run the same pipeline locally and pay only for the model calls it makes.
Paying a monthly subscription to turn a long interview into TikTok clips is one of the places a solo creator spends real money. A new open-source project called short-video-maker, from GitHub user mushigaite, does the same job locally and bills only the model calls it makes. The repository picked up 624 stars in the ten days since it was created on 5 October.
The README names OpusClip and Vidyo.ai as the paid tools it replaces. Paste a YouTube link, the tool downloads the video, writes a timestamped transcript with faster-whisper on your own machine, then asks an OpenAI, Gemini or MuAPI model to find the viral moments. It ranks each candidate on a scale of 0 to 100 against a list of hook moments, emotional peaks, opinion bombs, revelations, conflict, quotables, story peaks and practical value. Overlapping candidates are collapsed by score, the top N are kept, and each clip is rendered as a 9:16 vertical short with burned-in subtitles. There are no watermarks and no pre-clip credits.
What the tool does by itself
A pipeline the project describes in its own README has seven steps: download, transcribe, classify content type (podcast, interview, tutorial, vlog), rank highlights, dedupe, pick the top N, and auto-crop with an optional AI-generated hook sentence at the start. The content-type step tunes the virality prompt to the pacing of that kind of video. A --no-hook flag skips the opening sentence for creators who want to keep the clip untouched.
A built-in web interface queues several videos at once, with the API key still kept in a local .env file. The project offers its own HTTP API for creators who want to call it from another system.
The price and the licence
The tool is MIT-licensed. Gemini has a free tier with a daily limit; OpenAI and MuAPI charge per call. Whisper runs locally and is free. So the running cost is one model call per video for highlight ranking and one more for the optional hook, and nothing else.
A creator who was paying a short-video SaaS fifteen dollars or more a month gets a tool that charges only for the token work it does, and keeps the source video on their own machine instead of uploading it to a third party.
The traction
The project has 624 stars and 75 forks since its first commit on 5 October. 75 forks against 624 stars is a sign of real use rather than a padded star count, since clicking fork takes a developer who wants the code. The repository had a commit on 11 October, so it is maintained.
The limit a reader should see is that the tool is a wrapper around a hosted model. If the model picks poor clips, the output clips are poor. The ranking prompt is also a fixed template, so a creator who already judges short-form content well will do better than the default.
For a marketer or founder cutting demo videos, product talks or long interviews into vertical posts, the saving is one less monthly subscription and keeping the source video on their own hardware.
Source
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.