short-video-generator-AI

This turns a long YouTube video into vertical clips without you scrubbing a timeline. It downloads the source, transcribes locally with faster-whisper, classifies the content type, then scores every transcript segment against a virality rubric covering hooks, emotional peaks, and quotables. Overlapping candidates collapse and the top few render as 9:16 verticals, each able to open with an AI-written hook. It targets OpusClip, which charges a subscription for the same job.

You are not stuck in a terminal. There is a CLI, a local web page for queueing videos and setting flags visually, and an API for your own code. Transcription runs on your machine, so the only external cost is one LLM provider for classification and scoring. Gemini has a free tier with daily limits, MuAPI is pay per use, OpenAI is paid only.

Python 3.10 or newer and an env file. Whisper model size runs from tiny to large-v3 depending on your GPU. Solo creators and anyone repurposing a podcast are the audience. MIT, free, no watermarks, no per-clip credits.

The catch is that the headline promises more than the pipeline documents. Subtitles, translation, and voiceover are all advertised up top, but the seven documented steps end at cropping and the optional hook. Confirm they exist before building a workflow on them.