ClipForge

Tools that turn a long video into social media shorts, like OpusClip, are paid and billed monthly. I did not want a subscription, so I built ClipForge, my own fully local version, using Ollama to run the language models directly on my machine.

You hand the app a long video, either through a YouTube link or from a file already on the PC. It analyses the content, spots the interesting passages and automatically produces a series of ready-to-post vertical clips. On a 22 minute video, it generates 18 clips for example.

Each clip comes out finished: vertical reframing with face tracking so the speaker stays in frame, burned-in subtitles with keyword highlighting, silence trimming, plus a generated title, description and hashtags for publishing. Every clip gets a score so the most striking moments surface first.

I then added an optional cloud API for transcription, alongside the local pipeline. Local stays free and keeps videos private, while the API gives more accurate transcription, which improves both subtitle quality and the detection of the best moments. The user picks whichever matters most to them.

Processing leans on the machine's GPU to stay fast despite the workload. I built this one solo in early 2026, from the video analysis all the way to the interface.