SINGAPORE, Aug. 12, 2025 /PRNewswire/ — The SkyWork AI Technology Release Week officially kicked off on August 11. From August 11 to August 15, a new model will be unveiled each day, covering cutting-edge models for core multimodal AI scenarios.
A week ago, DeepMind released a major update to its interactive world model—Genie 3—enabling real-time, long-sequence generation. This advancement has drawn significant attention to world models. However, Genie 3 was not open-sourced, leaving the community to speculate about its implementation.
On August 12, Skywork unveiled an upgraded version of the self-developed Matrix series’ interactive world model—Matrix-Game 2.0. It also delivers interactive, real-time, long-sequence generation in general scenarios. To drive progress in interactive world modeling, Matrix-Game 2.0 has been fully open-sourced, marking the industry’s first open-source solution for real-time, long-sequence, interactive generation in general scenarios.
Matrix-Game 2.0 open source addresses:
- Technical report: https://github.com/SkyworkAI/Matrix-Game/blob/main/Matrix-Game-2/assets/pdf/report.pdf
- Project homepage: https://matrix-game-v2.github.io/
- HuggingFace: https://huggingface.co/Skywork/Matrix-Game-2.0
- GitHub: https://github.com/SkyworkAI/Matrix-Game
Matrix-Game 2.0 achieves a breakthrough in real-time generation and long-sequence handling. Compared to its predecessor, the 2.0 version prioritizes low-latency, high-frame-rate performance for extended interactions, enabling stable 25 FPS continuous video generation across complex scenes. Its generation length scales to minute-long sequences, drastically improving temporal coherence and real-world usability.
While delivering a significant boost in inference speed, Matrix-Game 2.0 maintains precise comprehension of physical laws and scene semantics. It enables users to freely explore, manipulate, and construct virtual environments in real time through simple instructions—yielding well-structured, detail-rich, and logically coherent virtual spaces.
With these capabilities, Matrix-Game 2.0 not only breaks down the barriers between content generation and interaction but also unlocks new possibilities for cutting-edge applications such as virtual humans, game engines, and embodied AI. It provides a robust technical foundation for building a universal virtual world.
Currently, Matrix-Game 2.0 boasts three core advantages:
High-frame-rate, real-time long-sequence generation: The model supports fluid movement (forward/backward, left/right) and camera/view rotation. Users can intuitively control characters in the scene via simple commands. The system generates seamless footage in real time at 25 FPS, enabling minute-long interactive sequences in a single session. Character movements are lifelike, smooth, and precisely responsive.
Cross-scenario generalization capability: The model demonstrates exceptional cross-domain adaptability. It is not only suitable for specific task scenarios but also supports simulations of diverse styles and environments—including urban, wilderness, and other spatial types, as well as realistic, oil-painting, and various visual styles.
Enhanced physical consistency: The model demonstrates a deeper understanding of physical rules. Characters generated by the model exhibit physically plausible movements when navigating complex terrains such as steps and obstacles, which improves immersion and controllability.
The open-source release of Matrix-Game for interactive video generation underscores Skywork’s strategic foresight in AI development. This initiative will accelerate development across Skywork’s multi-model AI ecosystem. Moving forward, Skywork remains committed to pioneering and open-sourcing advanced AI solutions. By collaborating with global developers and users, we aim to build next-generation platforms that accelerate the global advancement of AGI.
