Roblox Showcases Its Experimental AI Technology with a Mind-Bending Clair Obscur: Expedition 33 Inspired Creation

Your subscription is confirmed!

Get ready for your weekly dose of everything gaming, from the titles you adore to the ones you’ll soon be obsessed with, along with stories from the vibrant communities that bring them to life.

Dive into our exclusive GTA 6 newsletter, packed with the latest news, insider insights, and rumor breakdowns from our dedicated experts.

From the minds behind Edge comes a weekly newsletter focused on the gaming industry, featuring in-depth analysis from seasoned writers, professional guidance, and a glimpse into what’s coming next.

Tech enthusiasts, join our free newsletter for a weekly roundup of the latest innovations, cutting-edge gadgets, and much more.

Stay updated with our new Switch 2 newsletter, where we deliver the latest news and discussions about Nintendo’s upcoming console, along with game recommendations.

Subscribe for a weekly summary of essential movie and TV news, including trailers, interviews, reviews, and in-depth features.

Get exclusive previews, competitions, and details about special events every month!

Recently, the Roblox Corporation showcased its experimental generative AI technology, aiming to empower players to create entire game worlds. However, the initial results were less than stellar, with the tech producing something resembling Clair Obscur: Expedition 33—if you really squint and ignore its flaws.

Roblox thrives on user-generated content, so it’s no surprise that the company is exploring generative AI. In a series of recent tweets, they introduced a feature called 4D Generation, which allows users to create “interactive 3D objects like cars, plants, and more” through prompts.

Next up is ‘real-time dreaming,’ a more experimental concept currently in development. This feature promises users the ability to generate fully playable video worlds based on any text or image, which can then be converted to “Roblox native” for multiplayer experiences. In a twist of irony, the company refers to this concept as “Dream Theater,” where one user dreams while others watch and provide prompts.

Curious about what real-time dreaming can do right now? It’s capable of generating what could be a game of the year contender. Roblox released a video demonstrating the tech, where a prompt as simple as ‘woman in a glowing cave’ produced a scene reminiscent of Clair Obscur: Expedition 33’s Maelle navigating a blurry version of the acclaimed RPG’s Flying Waters area.

In our research lab, we are developing “real-time dreaming”—the ability to create fully playable video worlds from any text or image. Our real-time, action-conditioned world model (currently running internally at 16fps at 832x480p) is trained on a mix of data…

“Why did it create Expedition 33?” one viral tweet questions. “Expedition 33 looks terrible in this; how did it win game of the year?” another user remarks.

Roblox co-founder and CEO David Baszucki clarifies that the model was built using internal Roblox data alongside open-source video data. This internal data encompasses the 13 billion hours players spend in-game each month, which is now being utilized to train world models and AI-driven NPCs.

A former Square Enix executive commented on the viral Roblox hit Grow a Garden, describing it as “somewhat evil”: “Do you want to pay to grow this faster? Do you want to pay to steal this from someone else’s garden?”