plumblucker/clipforge

ClipForge is a set of PowerShell scripts that turn long stream recordings into short clips, running entirely on your own computer with your own GPU. Nothing is uploaded to a server, there is no subscription, and no account is required. It finds good moments in a recording three different ways. It can scan the audio track for loudness spikes, on the idea that your loudest reactions are usually your best moments, and produce a ranked shortlist you can review. It can transcribe the whole recording using faster-whisper, giving you a timestamped text file and subtitle file you or an AI can read through to pick moments by what was actually said. And it can watch for a specific spoken codeword you choose ahead of time, so saying a phrase like clip that during your stream marks that moment automatically, with the resulting clip starting slightly before you said it. Once you have picked a moment, a separate script performs a frame accurate cut of the video. It can produce a normal landscape clip or a vertical clip sized for Shorts, either by blurring the edges or cropping in to fill the vertical frame. A single command can also run the entire pipeline, loudness scan, transcription, and codeword detection, on a whole folder of recordings at once. The tool is built for Windows and needs PowerShell, ffmpeg on your system path, and Python with faster-whisper installed if you want transcription or codeword detection. An NVIDIA GPU is recommended for faster transcription, though it will still work on a CPU, just more slowly. The project describes itself as an honest shortlist and clipper rather than a fully automatic editor, meaning it points you toward good moments but leaves the final selection to you. It was created by a solo developer who got tired of manually scrubbing through long stream recordings, and it is released under the MIT license.