ViMax

ViMax is an open-source agentic multi-agent video generation framework from HKUDS (Hong Kong University Data Intelligence Lab). It integrates director, screenwriter, producer, and video generator into one workflow. Users feed in a concept, and ViMax autonomously handles scriptwriting, storyboarding, character-consistency management, reference image selection, camera planning, and final video generation — end-to-end pipeline fully open-sourced.

The framework ships four core workflows. Idea2Video takes a one-sentence concept and expands it into a multi-character, multi-scene video story. Novel2Video consumes novels or long-form text, automatically breaks them into scripts and storyboards. Script2Video accepts an existing screenplay and plans shots and camera rhythm. Multi-Camera Simulation emulates real-world multi-camera shooting while maintaining character placement and scene consistency.

Under the hood, multiple agents collaborate: ScriptAgent (screenplay breakdown), StoryboardAgent (storyboard design), ImageSelectionAgent (intelligent reference image selection), ImageGenerationAgent, and VideoGenerationAgent. The VideoGenerator backend is pluggable — Google Veo and other third-party models are supported out of the box. The ImageReference mechanism keeps characters and scenes consistent across long videos.

Use cases span film previsualization, novel/IP adaptation, marketing short-form video scripting, educational content production, and self-media creator rapid prototyping.

Deployment requires Python 3.12, the uv package manager, an OpenRouter API key, and at least one video generation API key (Google Veo recommended). Licensed under MIT, with 11.1k stars, 1.6k forks, and 364 commits on GitHub; last commit 2026-07-13, the project is actively maintained.