Imagine a world where artificial intelligence doesn't just chat or generate images—it builds entire virtual realms from scratch, revolutionizing how we create games, films, and even robotic simulations. But here's where it gets controversial: As AI pioneers like Fei-Fei Li push the boundaries, are we unlocking unprecedented creativity, or risking the erosion of human artistry in industries that have always relied on skilled labor? Dive into this exciting development as we explore how World Labs, the venture backed by AI visionary Fei-Fei Li, is shaking up the AI landscape with its groundbreaking debut product.
World Labs, the innovative startup spearheaded by Fei-Fei Li—who famously advocates for governments to adopt a practical, non-ideological stance toward AI advancements—is unveiling its inaugural commercial offering. Known as Marble, this tool is now accessible through a freemium model and premium subscriptions, empowering users to transform simple inputs like written descriptions, photographs, video clips, 3D floor plans, or panoramic views into fully editable and downloadable three-dimensional spaces. For beginners just getting into AI tech, think of it as a digital sculptor that takes your ideas and molds them into immersive environments you can tweak and save, much like how a 3D modeling program lets artists build virtual worlds, but with AI doing much of the heavy lifting.
This unveiling of the generative world model—initially teased in a restricted beta test just a couple of months back—follows a year since World Labs emerged from obscurity with a whopping $230 million in investment. This move positions the company at the forefront of the world model competition. To clarify for those new to the term, world models are AI frameworks that construct an internal map of a setting, enabling predictions of what might happen next and strategizing responses. Picture them as a smart simulator that learns how environments work, predicting outcomes like a chess AI foreseeing moves, but in three-dimensional spaces.
Rivals such as Decart, which crafts real-time playable simulations inspired by games like Minecraft, and Odyssey, offering streaming interactive 3D universes, have shared free prototypes. Even Google's Genie remains confined to limited research trials. What sets Marble apart—even from World Labs' own rapid model, RTFM—is its focus on creating stable, exportable 3D worlds instead of dynamically altering them during exploration. This approach, according to the team, minimizes glitches like unexpected changes or inconsistencies, allowing users to save creations as Gaussian splats (those are efficient 3D representations that compress data for faster rendering), meshes (structured 3D shapes), or video files. And this is the part most people miss: By prioritizing persistence, Marble ensures your virtual living room stays consistent, no matter how many times you revisit it, unlike fleeting, on-the-fly generations.
Marble stands out as the pioneer in integrating AI-powered editing tools and a blended 3D editor, enabling users to sketch out basic spatial frameworks—such as sketching walls or basic shapes—and then instruct the AI to fill in the visual flair. It's like providing the blueprint for a house and letting the AI decorate it with stunning details.
“This represents an entirely fresh type of model dedicated to crafting 3D worlds, and it's bound to evolve significantly. We've already made substantial enhancements,” remarked Justin Johnson, World Labs' co-founder, in an interview with TechCrunch.
Back in December, World Labs demonstrated how their prototype systems could produce explorable 3D scenes from just one photograph. Though captivating, these early versions had limitations: the scenes often felt like simplified animations, with restricted movement areas and sporadic visual hiccups. For instance, trying to walk around a generated forest might lead to abrupt ends or blurry edges.
During my hands-on experience with the beta, Marble effortlessly produced striking environments solely from image cues—ranging from fantasy game arenas to lifelike replicas of my own home. While borders sometimes shifted in the beta, the team assures that the official release has ironed out these issues. Interestingly, a scene I created in the beta using a single input turned out more polished and aligned better with my vision than the current version does with the same prompt. It's a reminder that AI improvements aren't always linear—sometimes, refinements introduce new challenges.
TechCrunch Event
San Francisco | October 13-15, 2026
I haven't had the chance to try the editing functions yet, but Johnson emphasizes their practicality for immediate applications in gaming, visual effects, and virtual reality endeavors.
“One of our core focuses for Marble is empowering user creativity,” Johnson explained. “You should always have a swift way to generate ideas, but also the option to go deeper and exert significant influence over your creations. We don't want the technology to overshadow human imagination.”
Marble's emphasis on user control begins with versatile input options. The beta was limited to single photos, requiring the AI to extrapolate hidden aspects for a complete 360-degree perspective. Now, with the full rollout, individuals can submit multiple photos or brief videos capturing a space from various viewpoints, resulting in highly accurate virtual duplicates—think of it as creating a digital twin of your office that's so precise, it could fool a visitor.
Then there's Chisel, an innovative 3D editing suite that allows users to outline rough spatial designs (like positioning walls or geometric shapes) and overlay text instructions for aesthetic guidance. Marble then renders the scene, separating design from appearance—akin to how web development uses HTML for structure and CSS for styling. But unlike purely text-driven edits, Chisel enables direct object manipulation.
“I can simply select the 3D element representing the sofa and relocate it,” Johnson illustrated.
An additional feature enhancing customization is the capacity to enlarge a world.
“After creating a world, you can extend it once,” Johnson noted. “As you approach areas that begin to degrade, you can instruct the model to expand or generate additional content nearby, adding finer details in those spots.”
For those aiming to build vast expanses, the “composer mode” allows merging multiple worlds. Johnson showcased this by fusing two pre-built environments: one a whimsical chamber of cheese with grape furnishings, and another a high-tech space station boardroom, creating a seamless, expansive hybrid.
The Journey Toward Spatial Intelligence
Marble comes in four pricing levels: Free (limited to four generations from text, images, or panoramas), Standard ($20/month for 12 creations, plus multi-input support and enhanced editing), Pro ($35/month with 25 generations, world expansion, and commercial usage rights), and Max ($95/month offering all functionalities and 75 generations).
Johnson anticipates initial applications in gaming, cinematic visual effects, and virtual reality.
Yet, the gaming community harbors divided sentiments about this technology. A recent survey from the Game Developers Conference revealed that roughly one-third of developers view generative AI as detrimental to the industry—up 12% from the previous year. Key worries include infringement on intellectual property, high energy demands, and a potential decline in content quality due to AI shortcuts. Last year, a Wired exposé uncovered how firms like Activision Blizzard employ AI to streamline processes and address workforce reductions, raising questions about whether this tech is a tool for innovation or a shortcut that devalues skilled craftsmanship. But here's where it gets controversial: Is AI in gaming a democratizing force, letting indie creators compete with giants, or is it unfairly displacing artists and designers? And this is the part most people miss: While AI can accelerate production, it might stifle the unique storytelling that human creators bring to games.
In the gaming realm, Johnson envisions developers leveraging Marble to produce ambient backgrounds and secondary settings, then importing these into platforms like Unity or Unreal Engine to incorporate interactivity, scripting, and programming.
“Marble isn't meant to overhaul the whole game development workflow, but to provide ready-made elements you can integrate,” he clarified.
For visual effects in movies, Marble addresses the shortcomings of AI video tools, such as erratic scenes and limited camera control, per Johnson. Its 3D assets enable filmmakers to arrange shots and adjust camera angles with pinpoint accuracy, like having a virtual set that's perfectly controllable.
Though World Labs isn't currently prioritizing virtual reality, Johnson points out the sector's desperate need for fresh material and buzz surrounding the launch. Marble already supports devices like the Vision Pro and Quest 3, with all worlds VR-ready for immediate viewing.
Marble could also extend to robotics. Johnson highlights that, unlike image or video creation, robotics lacks abundant training data. However, tools like Marble simplify creating simulated training grounds, helping robots learn in virtual environments before real-world tests—imagine training a delivery drone in a simulated cityscape to avoid obstacles.
Per a recent essay from Fei-Fei Li, World Labs' CEO and co-founder, Marble marks the initial stride toward a genuinely spatially aware world model.
Li posits that “upcoming world models will empower machines with spatial intelligence at a revolutionary scale.” Just as language models enable reading and writing, she envisions Marble-like systems fostering vision and construction abilities. Grasping how objects occupy and influence 3D spaces could spur innovations not just in entertainment and robotics, but in fields like scientific research and healthcare—think AI designing medical simulations for surgery training or modeling molecular structures for drug discovery.
“The fulfillment of our aspirations for intelligent machines hinges on spatial intelligence,” Li asserted.
What do you think? Does accelerating AI in creative fields like this excite you as a step toward boundless innovation, or do you worry it's undermining human jobs and originality? Share your views in the comments—do you agree with Li's vision, or see potential pitfalls we haven't addressed?
Have a sensitive lead or confidential files? We're delving into the AI sector's inner dynamics—from the corporations molding its future to the individuals affected by their choices. Contact Rebecca Bellan at rebecca.bellan@techcrunch.com or Russell Brandom at russell.brandom@techcrunch.com. For secure chats, reach them via Signal at @rebeccabellan.491 and russellbrandom.49.
