Roblox experiments with AI "real-time dreaming" to change game worlds on the fly
Less than a week after Google revealed Project Genie — an AI world model that generates interactive 3D experiences from prompts — Roblox shared early ideas for a similar concept it calls "real-time dreaming." The company wants creators to use AI world models and natural-language prompts to generate and modify experiences in real time.
What Roblox demonstrated
In a virtual briefing, Roblox SVP of engineering Anupam Singh showed The Verge a prerecorded demo of real-time dreaming. The short clip pictured a Viking-themed world that responded to live prompts: the AI added a tsunami wave, then generated a boat for the Viking character to board. Roblox CEO Dave Baszucki has shared what appears to be the same demo in a public video.
"If someone can dream it, they should be able to bring it to life," Singh wrote in a blog post about AI-driven creation on Roblox.
Where the tech stands
- Research stage: Roblox describes real-time dreaming as experimental. There is no timeline for public release, Karun Channa, senior director of product, told The Verge.
- Limited firsthand testing: Reporters were shown prerecorded demos but could not try the system themselves. Early tests of comparable tools, like Project Genie, produced mixed results.
- Quality questions: Real-time prompts can change a scene on the fly, but it remains unclear whether those changes will always be meaningful or enhance gameplay.
How Roblox sees AI fitting into game development
Roblox does not expect AI world models to fully replace human creators. Singh says making a great game still requires human creativity and a creative mindset. Similarly, other industry leaders have echoed that AI will augment but not supplant developers.
New "4D creation" tools in open beta
Alongside research into real-time dreaming, Roblox is launching "4D creation" tools that let developers integrate , interactive objects into live games. These tools were in closed beta and are now available in open beta. Players can prompt the AI to create items they can drive, fly, or shoot.
An existing experience called Wish Master demonstrates these tools. Developers pick a model from a menu, similar to choosing a model in an AI chatbot, then generate in-game objects. In Wish Master the world acts as an open sandbox for players to spawn and test creations.
Early impressions and limitations
- Novelty factor: Generating objects felt interesting at first, but the novelty wore off for some testers.
- Limited interactivity: In Wish Master, the environment offered little beyond open space, so players had few structured ways to use or integrate generated objects into longer experiences.
- Industry pushback: Some creators and game developers oppose AI tools, citing creative, ethical, and legal concerns. Several AI companies have faced copyright lawsuits, and a subset of developers choose not to use generative AI at all.
Why this matters in gaming news
AI-driven world models could change how games are built and experienced. For creators, natural-language prompts may speed iteration, debugging, and collaboration. For players, dynamic, on-the-fly world changes could enable new forms of emergent gameplay.
However, adoption depends on whether AI can reliably produce high-quality, useful content without undermining the creative role of human developers. As gaming news continues to cover advances from Google, Meta, xAI, and others, Roblox's experiments are a key signal that major platforms are betting on AI-assisted creation.
For now, real-time dreaming remains an intriguing research direction rather than a finished product. Developers and players should expect more experiments, open betas like 4D creation, and continued debate about the role of generative AI in game development.
— Jay Peters
Comments
No comments
Add Comment