poseljacob/agentic-video-editor

This project is a command line tool that turns raw video footage into a finished advertisement using a team of AI agents. You point it at a folder of video clips, describe what you want in a short creative brief such as the product, the target audience, the tone, and the desired length, and the tool handles scene detection, shot selection, editing, and quality review on its own. The work is split among four agents. A Director agent, powered by Google Gemini, searches through an index of your footage and produces an edit plan: it chooses which shots to use, decides how to order them, and picks where to place text overlays. A Trim Refiner agent then tightens the start and end points of each cut. An Editor agent renders that plan into an actual video file using a tool called FFmpeg. Finally, a Reviewer agent watches the finished video and scores it on five dimensions, including how well it matches the brief, pacing, visual quality, and overall watchability. If the score falls below a set threshold, the plan goes back to the Director for another pass, and each retry is saved as a separate version so you can compare them. Editing behavior is controlled through YAML files. Pipeline files define which agents run and in what order, plus retry rules. Style files describe pacing, segment timing, and text overlay placement for a particular kind of ad, such as a thirty second testimonial style commercial. To use it you need Python 3.11 or newer, FFmpeg installed on your computer, and a Google AI API key, since the agents run on Google Gemini. The tool is installed by cloning the repository and installing its Python dependencies with pip or the uv package manager. There is also an early, unfinished web interface called AVE Studio, built with FastAPI and Next.js, but the readme is explicit that it is not yet the recommended way to use the project and that the command line tool is the supported path. The project is released under the MIT license.