What is ai-video-agent?

AI Video Agent is a web application that turns a short text description into a short video. You type a topic or a brief, and the tool generates a script, creates visual slides, adds spoken narration, syncs subtitles, and lets you export the result as an MP4 file. It is built with Next.js and runs locally on your machine. The tool supports two production modes. In image mode, it generates a series of still images for each section of the script, pairs each one with text-to-speech audio, and assembles them into a slideshow-style video. In HTML mode, it generates animated web pages using HTML, CSS, and JavaScript for each section instead of static images. HTML mode is better suited for technical demonstrations and data visualizations where motion adds clarity. Three visual styles are available for HTML mode: a dark tech look with blue highlights, a terminal and matrix-style aesthetic with green text, and a warm soft palette for personal or storytelling content. You can regenerate individual sections without remaking the entire video, and the tool saves progress so a long job can resume if interrupted. To run the project you need Node.js 22 or newer, a MySQL database, and API keys for an OpenAI-compatible text model, a text-to-speech service, and an image generation service. Configuration is done through a single environment file. A Docker Compose file is included so you can start a local MySQL instance without installing it manually. The README describes the project as experimental and notes it was built with assistance from an AI model called MiniMax-M3. Known bugs may exist. The code is open source.