gargvr/shorts-autopilot

Shorts Autopilot is a Python tool that automatically finds long-form YouTube videos in a topic area, cuts the best moments into vertical short clips, adds animated captions, and uploads them to your YouTube channel on a schedule. The goal is to run a YouTube Shorts channel that publishes content without ongoing manual work. Each run follows a fixed pipeline. It searches YouTube for recent popular videos matching keywords you define in a YAML config file. It picks one video you have not clipped before, downloads it, and then runs one of two cutting strategies: for talk-style or interview content, it transcribes the audio with Whisper and uses GPT-4o to identify the strongest self-contained moments, then crops each to 9:16 (vertical) framing and adds word-by-word captions. For sports or highlight content, it finds audio peaks like crowd noise and renders them as vertical clips with the original commentary. After cutting, it generates ready-to-paste titles, captions, and hashtags for each clip, then uploads them to YouTube as private by default. A local ledger file remembers which source videos have already been processed, so each run automatically moves to something new. You configure the niche, keyword list, privacy setting, and pipeline type in a single YAML file. A dry-run flag lets you preview what would happen without downloading or uploading anything. The only paid costs are for the talking pipeline: OpenAI Whisper transcription at roughly $0.006 per source minute and a single GPT-4o call for moment selection, totaling under $0.20 for a typical 8 to 12 minute video. The highlights pipeline has no API costs beyond the YouTube Data API. A local Whisper option (faster-whisper) is available to eliminate the transcription cost entirely. The tool runs on cron, Windows Task Scheduler, or a Claude Cowork scheduled task. Uploads default to private, and the README includes a clear legal notice: reuploading copyrighted content violates YouTube terms of service regardless of attribution. The code is MIT licensed, you are responsible for what you run through it.