ViMax is an open-source agentic video-generation framework from the HKUDS research group that splits filmmaking into specialized AI roles — director, screenwriter, producer and video generator — and orchestrates them into one automated pipeline. Released under the MIT license, it has attracted more than 12,700 stars on GitHub and was actively developed in 2026.
Text-to-video models produce striking clips but struggle with coherent, multi-scene stories. ViMax attacks that gap with agents: a screenwriter agent develops the narrative and shot list, a director agent plans scenes and camera work, a producer agent coordinates assets and consistency, and a video-generator agent renders the footage. By chaining these roles, the framework aims for videos that hold characters, style and storyline together across shots, rather than a single disconnected clip. The repository includes the full pipeline so researchers and creators can run, study and extend each stage.
As a research framework, ViMax is also a useful lens on where AI filmmaking is heading: the division of labor mirrors how human productions scale creativity, and each agent can improve independently as better language, image and video models appear. Experiments in the project demonstrate story-driven outputs — dialogued scenes and planned sequences — that a single prompt to a video model cannot reliably produce today.
ViMax is free and open source. Running it requires your own compute and access to the underlying generation models you plug into the pipeline.
ViMax is aimed at AI-video researchers, creative technologists and indie creators experimenting with automated storytelling — anyone who wants to move beyond single-prompt clips toward directed, multi-scene AI video production.