AI NewsOpen sourceAnnouncement

neonclipping turns long videos into captioned 9:16 clips, locally

Ajay K Meena published neonclipping, an open-source tool that turns a YouTube link into captioned 9:16 short clips. It has 494 GitHub stars after one day.

AI News

Editorial2 min read

LinkedInX
GitHub social card for the neonclipping repository

Image: GitHub

Why it mattersPaid short-clip services charge a subscription and upload your video to their servers. A free local tool lets a solo creator test the format for nothing first.

A cinematographer in Jaipur, Ajay K Meena, published neonclipping on 2026-11-19. He tagged its first release the same day. The tool takes a long YouTube, Twitch or Kick link, or an MP4 file on your disk. It returns short vertical clips with burned-in captions. The whole pipeline runs on your own computer. The repository had 494 stars at the time of reading.

The tool is MIT licensed and written in Python. The author runs it on Windows with a one-click batch file. Linux and macOS use a shell script. The user interface is a web page the tool serves on 127.0.0.1, port 8770. The server never opens to the network.

How it picks moments

neonclipping scores moments with four signals from the source video. Each one is listed in the project's README. Loudness jumps against a two-minute rolling baseline. Energy between two and four kilohertz, which the author says catches a raised voice. Quiet-to-loud transitions. The fourth signal only works on YouTube links with captions. A local language model scores the caption text for emotional content and narrative payoff.

A deeper mode downloads the full video and adds motion detection. It uses OpenCV's Farneback optical flow. It evaluates four times more candidate moments than the fast mode. Fast mode is the default for videos over ten minutes. Fast mode uses only the audio and existing captions, so it never downloads the video.

What the author measures

The README gives one measured example. An 85-minute YouTube stream, using YouTube's own captions, produces three clips in about 97 seconds. That is the fast mode. The deep mode is slower and the README does not state by how much.

What is bundled

The tool uses FFmpeg to encode video and yt-dlp to download it. It uses faster-whisper to transcribe when there are no captions. It uses Ollama with a local language model to score moments and write titles. The author suggests llama3.2:3b. It uses OpenCV YuNet to detect the speaker's face. It uses Streamlink to record live streams. Each is an existing open-source project. neonclipping is the orchestration and the rules.

Clips come out at 1080 by 1920 pixels, with burned-in captions and a title line, as MP4 files. The user can rate each clip from one to five stars. The tool uses those ratings to adjust how it scores the next video.

What this means for a small team

A solo creator today pays a monthly subscription to a hosted service to cut short clips from a stream. The service takes the video, picks moments and returns MP4s. That is a working business and nothing here changes it. What changes is the cheapest option. A person can now test the format on their own footage, with their own captions and their own face framing. The cost is only what it takes to run Ollama on a laptop. If the output is good enough, they keep it. If not, they go back to the paid service without having paid for the test.

Source

iamajmeena/neonclipping on GitHub and the v1.0.0 release.

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.

Share
LinkedInX
Start a project