# Visko Unveils Orbis: A Real-Time, Interactive AI Video Generation Platform
## A New Era in Generative Video
A California-based artificial intelligence company has launched a publicly accessible foundation model that promises to redefine how video is created using AI. Unlike traditional generative video tools that produce fixed-length clips in a single batch, the new model generates worlds continuously, streaming them in real time with the ability to respond to user inputs on the fly.
The company, Visko Platform Inc., announced that its model — called Orbis — is now open for public use at visko.ai. In conjunction with the launch, Visko disclosed that it has secured $10 million in pre-seed funding, with the round led by Llama Ventures.
## How Orbis Differs from Conventional Video AI
Most AI video generation systems work by processing every frame of a clip simultaneously and presenting the final output only once the entire sequence has been computed. Orbis takes a fundamentally different approach: it completes the first frame, streams it immediately, and then generates each subsequent frame based on what has already been produced. This is conceptually similar to how large language models predict the next token in a sequence.
This design means that the world being generated maintains persistent memory of what came before. A user can change directions, adjust the scene, or introduce new elements at any point during generation, and the model seamlessly incorporates those changes without restarting or losing coherence.
## Technical Capabilities
Orbis is engineered to deliver 4K resolution video at a steady 24 frames per second. One of its most notable claims is the ability to sustain generation for hour-long durations without any noticeable degradation in quality, color accuracy, or visual consistency over time.
Under the hood, Orbis relies on a latent world model that evaluates candidate future states during the generation process. This scoring mechanism helps steer the output toward motion and object interactions that are grounded in real-world physics, ensuring that the generated content behaves in ways that look natural and plausible.
“Our core innovation is the memory architecture,” said Qing (Will) Yin, founder and CEO of Visko. “We embed physical dynamics directly into each frame as it is produced, which actually reduces the computational overhead compared to traditional approaches. This makes long-form egocentric video generation feasible without any interruptions.”
## About Visko Platform Inc.
Visko was established in 2025 in Sunnyvale, California. Yin, a Stanford University Ph.D. graduate in computational mathematics and mechanics, spent three years conducting research at Apple before founding the company. The startup currently employs a team of 16 individuals whose collective experience spans major technology organizations including Apple, Google DeepMind, Meta, Amazon, and Tesla. Visko is also collaborating with Google on certain initiatives.
## Potential Applications
The ability to generate interactive, physics-consistent video in real time opens up a wide range of practical use cases. One of the most immediately relevant is in the field of robotics, where Orbis could be used to produce synthetic training footage for robots operating in scenarios that would be too dangerous, expensive, or logistically difficult to film in the real world.
Another promising application is in autonomous vehicle development, where the model could simulate rare and dangerous driving edge cases, helping engineers train safety systems without exposing real vehicles or passengers to risk.
Beyond robotics, advisors and industry observers see potential in gaming, live commerce, education, real-time creative media, and physical simulation. Michael I. Jordan, chairman of Visko’s advisory board and a professor at the University of California, Berkeley, described the technology as one of the most ambitious directions currently being pursued in video AI.
## Areas Where Further Development Is Needed
Visko’s technical report also acknowledges that Orbis still has room for improvement. The company has identified several dimensions where its model underperforms compared to competing systems. Yin highlighted the need for more sophisticated memory architecture to maintain even stronger consistency across long generation sessions.
Additionally, Visko is working on enriching the action descriptions used in training data. Currently, the model can interpret high-level commands directed at robotic agents, but finer-grained instructions — such as specifying exact angles of movement — are not yet fully integrated. On the engineering side, scaling the model’s parameters and training data remains a priority. The team is also exploring how to incorporate data from additional sensor modalities, such as tactile sensors, to further enhance realism.
## Advisory Board and Distribution
Visko’s advisory team includes prominent academics such as Steve WaiChing Sun of Columbia University and Mengye Ren of New York University, alongside Jordan. The company’s distribution is managed through a partnership with Reactor.
## FAQ
**What is Orbis?**
Orbis is a real-time, interactive video generation foundation model developed by Visko Platform Inc. It streams video continuously, maintains persistent memory of the scene, and allows users to modify prompts while generation is in progress.
**How does Orbis differ from other AI video generators?**
Unlike conventional tools that generate an entire clip at once and present nothing until completion, Orbis produces and streams each frame sequentially, using a latent world model to ensure physical plausibility and narrative coherence over time.
**What resolution and frame rate does Orbis support?**
Orbis delivers 4K video at 24 frames per second in real time.
**How long can Orbis generate video continuously?**
According to Visko, Orbis can sustain hour-scale generation without experiencing quality degradation or color drift.
**Who founded Visko?**
Visko was founded by Qing (Will) Yin, a Stanford Ph.D. in computational mathematics and mechanics who previously worked as a researcher at Apple.
**What is Visko’s team size and background?**
Visko has a 16-person team composed of alumni from Apple, Google DeepMind, Meta, Amazon, and Tesla.
**How much funding has Visko raised?**
Visko closed a $10 million pre-seed funding round led by Llama Ventures.
**What are the primary use cases for Orbis?**
Key applications include robotic training simulation, autonomous vehicle edge-case testing, gaming, live commerce, education, and real-time creative media.
**Is Orbis available to the public?**
Yes, Orbis is now open for public access at visko.ai.
**Are there known limitations?**
Yes. Visko acknowledges that the model needs improvements in memory consistency, finer action description capabilities, and scaling of model and data size. The company is also exploring integration with additional sensor types such as tactile data.
## Conclusion
The introduction of Orbis marks a significant step forward in the evolution of AI-generated video. By shifting from static, one-shot clip generation to a continuous, interactive, and physics-aware streaming paradigm, Visko has opened the door to applications that were previously impractical or impossible. From training robots in simulated environments to creating immersive real-time visual experiences, Orbis represents a new class of generative AI technology — one that runs not as a finished product, but as a living, responsive process.
As Visko continues to refine its architecture, expand its training data, and integrate multi-modal sensor inputs, the potential impact across robotics, automotive safety, entertainment, and education is substantial. The company’s progress in its pre-seed phase suggests that the convergence of real-time generation and physical simulation is moving closer to mainstream adoption.
Thank you for reading
