The Future of Real-Time World-Building: Visko and Google Cloud to Host "Live Models Fireside Chat"
The intersection of artificial intelligence, high-performance computing, and interactive media is rapidly approaching a paradigm shift. On Wednesday, October 14, in Los Angeles, AI research firm Visko and Google Cloud will convene industry professionals for an exclusive "Live Models Fireside Chat." This event promises to peel back the curtain on the next generation of AI-generated environments—systems capable of generating, sustaining, and evolving complex, physics-grounded worlds in real time.
As the creative industries—spanning animation, visual effects (VFX), and game development—grapple with the integration of generative AI, the focus is shifting from static, prompt-based image generation to dynamic, persistent environments. The upcoming fireside chat aims to bridge the gap between creative ambition and the formidable computing infrastructure required to power these "Live Models."
Main Facts: Redefining Virtual Environments
The core of the discussion centers on Visko’s proprietary technology, Orbis. Unlike conventional generative AI, which often struggles with temporal consistency—the ability to keep an object or environment looking the same across different frames or interactions—Orbis is a foundation model designed for persistence.
What are "Live Models"?
Visko defines "Live Models" as AI-generated systems that prioritize three critical pillars:
- Continuous Interaction: Environments that do not merely react to a single prompt but exist as a living, breathing space that updates based on ongoing user input.
- Persistent State: The ability of the environment to "remember" previous states, ensuring that if a user alters a landscape or moves an object, those changes remain consistent as the user continues to explore.
- Physics-Grounded Generation: Moving beyond mere visual synthesis, these environments are governed by physical rules, ensuring that interaction feels intuitive and logically sound.
For the creative sector, this represents a significant departure from traditional production pipelines. Instead of pre-rendering assets or hand-crafting every polygon in an engine, creators may soon be able to "direct" an environment that constructs itself in real-time.
Chronology: The Road to Real-Time Synthesis
The journey toward real-time, persistent AI world-building has been a multi-year effort, accelerating significantly in the last 18 months.
- Pre-2022: The industry was largely focused on static image generation (GANs and early diffusion models). Continuity was a secondary concern, and real-time interaction was largely impossible due to latency.
- 2023: The rise of Large Language Models (LLMs) and improved diffusion architectures began to show promise in spatial understanding. Researchers began exploring "World Models"—AI systems that can simulate the physical world.
- Early 2024: Visko’s research into Orbis began to gain traction, specifically targeting the limitations of existing generative models in the context of gaming and film production.
- October 2024: The partnership with Google Cloud marks a pivot toward infrastructure. The realization that "Live Models" are only as good as the hardware that runs them has brought compute giants like Google to the forefront of creative tool development.
Supporting Data: The Infrastructure Challenge
Running a generative model is computationally expensive. Running a generative model in real-time while maintaining physics-based persistence is a feat of engineering that requires massive parallel processing power.
The Role of Tensor Processing Units (TPUs)
A significant portion of the fireside chat will be dedicated to Google Cloud’s Tensor Processing Units (TPUs). While many in the creative industry are familiar with GPUs (Graphics Processing Units), TPUs are custom-designed Application-Specific Integrated Circuits (ASICs) built by Google specifically for machine learning workloads.
- Inference Efficiency: While training a model is a massive, one-time investment, "inference"—the act of using the model to generate content—happens continuously in a Live Model system. TPUs are optimized to handle these constant, high-volume requests with lower latency than traditional hardware.
- Scaling: As worlds become more complex, the demand for bandwidth and compute memory increases exponentially. Google’s infrastructure allows Visko to scale these environments across distributed networks, potentially allowing multiple users to inhabit the same AI-generated space simultaneously.
The integration of TPUs is not just a technical detail; it is the enabler that makes the "Live" in "Live Models" possible. Without this level of specialized hardware, the frame rates required for immersive virtual production would be unattainable.
Official Responses and Expert Perspectives
Visko CEO Will Yin is expected to lead the discussion on why current AI systems are insufficient for the needs of modern filmmakers and developers. "The current iteration of AI is often treated like a printer—you give it a prompt, and it spits out a result," Yin has noted in preliminary briefings. "Live Models are not printers; they are environments. They require memory, temporal consistency, and a foundation of physics that standard models simply aren’t built to maintain."
Dave Elliott, head of developer advocacy for AI at Google Cloud, will provide the technical counterpoint. Elliott’s presence signals that Google is positioning its cloud infrastructure not just as a host for data, but as a creative engine for the next generation of entertainment. "We aren’t just talking about hosting files," Elliott notes. "We are talking about hosting the active compute processes that create reality in real-time."
Implications for the Creative Industry
The shift toward Live Models has profound implications for the future of entertainment and professional workflows.
1. Previs and Virtual Production
Virtual production—popularized by shows like The Mandalorian—relies on high-end game engines like Unreal Engine. While powerful, these engines require massive teams to build assets. With Orbis and similar technologies, a director could potentially describe a set, and the AI would generate a navigable, persistent environment in real-time, allowing for rapid iteration without the need for traditional 3D modeling bottlenecks.
2. The Evolution of Gaming
The gaming industry is arguably the most affected. We are moving toward a future where "non-playable characters" (NPCs) and environments are not pre-scripted. If an environment can be generated and sustained in real-time, the "game world" could theoretically be infinite, evolving based on the player’s unique choices rather than a developer’s predetermined map.
3. Ethical and Creative Concerns
Despite the excitement, the rise of AI-generated worlds brings concerns regarding authorship and job displacement. If an AI can generate a world, what happens to the concept artists, environment designers, and lighting technicians? The industry will need to navigate how these tools serve as an extension of human creativity rather than a replacement for it.
Event Details and Registration
The "Live Models Fireside Chat" is an invitation-only event, reflecting the specialized nature of the discussion.
- When: Wednesday, October 14, 6:30 PM – 9:00 PM (Pacific Time).
- Where: Los Angeles, CA (Venue details provided upon registration approval).
- Format: Fireside chat, panel discussion, and networking.
- Registration: Due to limited capacity, attendance is subject to host approval. Interested professionals in animation, VFX, and game development are encouraged to register here.
As the barriers between the digital and physical worlds continue to blur, events like this serve as essential touchpoints for those looking to shape the future of visual media. Whether one views these developments with excitement or trepidation, the emergence of persistent, real-time, AI-generated environments is a development that the creative world can no longer afford to ignore.
