What's Happening?
Visko Platform Inc. has officially launched Orbis, a new foundation model designed for continuous, real-time video generation. Unlike conventional AI video generation that produces a static clip, Orbis streams video live and allows users to modify prompts
instantly, with updates reflected in real-time. The company describes Orbis as a 'Live Model' that runs continuously with persistent memory, interactivity, and physics-grounded generation for unlimited durations. Visko has also announced the successful closure of a $10 million pre-seed funding round, led by Llama Ventures. Qing (Will) Yin, founder and CEO of Visko, stated that Orbis addresses the challenge of maintaining consistency in long-running AI-generated worlds, ensuring physical, visual, and narrative coherence. The model is capable of delivering 4K video at 24 frames per second and can sustain hour-scale generation without quality degradation.
Why It's Important?
The introduction of Visko's Orbis model represents a significant advancement in AI-driven content creation and simulation, moving beyond static video generation to dynamic, interactive, and continuous streaming. This capability has profound implications for various U.S. industries, particularly in areas requiring realistic and adaptable virtual environments. For instance, in robotics, Orbis can provide interactive training environments for autonomous systems, allowing for the simulation of unsafe or expensive scenarios. The gaming industry could leverage this technology for more immersive and dynamic virtual worlds, while creative media could benefit from real-time, steerable video generation. The ability to change prompts mid-run and have the world update instantly offers unprecedented flexibility for developers and content creators. The $10 million pre-seed funding also signals investor confidence in this innovative approach, potentially fueling further research and development in real-time AI models and their applications across diverse sectors.
What's Next?
Visko plans to continue refining the Orbis model, with ongoing research focused on improving the memory structure to enhance video consistency across generations. The company also aims to incorporate more detailed action descriptions into its data collection for training, enabling more precise control over AI-generated actions, such as robotic movements. On the engineering front, Visko intends to scale up the model in terms of both model size and data, currently focusing on audio and video data while exploring the injection of other sensor data like tactile information. The company is actively seeking partnerships with hardware companies specializing in robotic sensors and applications. Advisors, including Michael I. Jordan of the University of California, Berkeley, anticipate a broad range of commercial opportunities for Orbis, from robotics and physical simulation to gaming, live commerce, education, and real-time creative media. Orbis is also expected to contribute to creating environments for testing corner cases for safer autonomous vehicles and humanoid robots.
Beyond the Headlines
The development of 'Live Models' like Orbis signifies a deeper shift in how AI interacts with and simulates reality. By continuously generating and streaming worlds with persistent memory and physics-grounded dynamics, Orbis moves closer to creating truly autonomous and adaptive virtual environments. This capability raises ethical considerations regarding the distinction between simulated and real experiences, particularly as these models become more sophisticated and indistinguishable from reality. The potential for Orbis to create training footage for scenarios that are unsafe or expensive to stage highlights its value in risk mitigation and safety protocols across various high-stakes industries. Furthermore, the emphasis on physical plausibility and consistent narrative in AI-generated worlds could lead to new standards for realism and immersion in digital content, potentially blurring the lines between digital and physical experiences and influencing how humans perceive and interact with AI-created realities.
