Roblox launches 4D creation AI tool in open beta, enabling interactive object generation

Roblox's 4D creation feature is now available in open beta | TechCrunch

Last year, Roblox launched an open-source AI model that could generate 3D objects on the platform, helping users quickly design digital items such as furniture, vehicles, and accessories. The company claims the tool, called Cube 3D, has so far helped users generate over 1.8 million 3D objects since it was rolled out last March. On Tuesday, the company launched the open beta for its anticipated 4D creation feature that lets creators make not just static 3D models, but fully functional and interactive objects. The feature has been in early access since November. Roblox says 4D creation adds an important new layer: interactivity. With this technology, users can design items that can move and react to players in the game. At launch, there are two types of object templates (called schemas) that creators can try out. The first is the "Car-5" schema, which is used to create a car made of five separate parts: the main body and four wheels. Previously, cars were a single, solid 3D object that couldn't move. The new system breaks down objects into parts and assigns behaviors to each so that they function individually within the virtual world. The AI therefore can generate cars with spinning wheels, making them more realistic and interactive. The second is called "Body-1," which can generate any object made from a single piece, like a simple box or sculpture. The first experience with 4D generation is a game called Wish Master, where players can generate cars they can drive, planes they can fly, and even dragons. In the future, Roblox plans to let creators make their own schemas so they'll have more freedom to define how objects behave. The company says it is also developing new technology that could use a reference image to create a detailed 3D model that matches the image's style (example below.) The company says it is developing more ways to help people create games and experiences using AI, including a project it has dubbed "real-time dreaming." Roblox CEO David Baszucki last month explained that this project would let creators build new worlds using "keyboard navigation and sharing real-time text prompts." The open beta comes on the heels of Roblox's recent implementation of mandatory facial verification for users to access chat features in the game, following lawsuits and investigations related to child safety.

Roblox calls its take on AI world models 'real-time dreaming'

Less than a week after Google showed off Project Genie, a tool powered by an AI world model that lets users generate interactive 3D experiences from prompts, Roblox is sharing some early ideas about how it wants to use AI world models and prompts to let creators generate experiences and change them in real time -- something it calls "real-time dreaming." Tech companies, including Google, Meta, and xAI, have been increasingly interested as of late in the idea that AI can help users make interactive experiences -- even though creators of many kinds are pushing back against AI tools, a growing number of video game developers think generative AI is bad for the gaming industry, and some developers are choosing not to use AI development tools at all. Some companies that make AI tools, including Google, have also been subject to copyright infringement lawsuits. In a virtual briefing, Roblox SVP of engineering Anupam Singh showed The Verge an example of what real-time dreaming might look like using a prerecorded video. In the short demo, an AI-generated viking-themed world responded in real-time to prompts that added a tsunami wave and then a boat for the viking character to board. You can see what appears to be the same demo I was shown in this video featuring Roblox CEO Dave Baszucki. Roblox's real-time dreaming is still in a "research stage," Karun Channa, senior director of product, tells The Verge, and there's no timeline for when the tools will be available. I couldn't try real-time dreaming myself, so I can't vouch for how well the experience may actually work right now in practice. The experiences I made with Project Genie last week weren't very good, and even though Roblox's demo showed off the real-time changes (which Project Genie isn't capable of), I'm still skeptical that the ability to change the experiences on the fly with prompts will make a meaningful difference in how enjoyable they will be. "The next frontier of creation on Roblox is the continued AI-driven evolution of our creation platform that will allow creators to generate immersive environments, iterate, debug, and collaborate with their teams all through natural language prompts," Singh says in a blog post. "If someone can dream it, they should be able to bring it to life." But Roblox doesn't envision these types of AI world model tools completely taking over from traditional game development. Singh tells The Verge that making a game requires a creative mindset, and "we don't see a model replacing that creative part." (A Take-Two executive said something similar yesterday on an earnings call.) Today, Roblox is also launching "4D creation" tools developers can integrate into their games so that players can prompt AI to generate objects they can interact with. Previously, those tools had been in a closed beta, but now they'll be in open beta, and they'll let you generate things that you drive, fly, or shoot. You can get an idea of what developers can do with the tools in an experience called Wish Master, which has already implemented them. To use the new "4D" tools, you have to select them from a model picker menu like what you might find in a more traditional AI chatbot. After a few minutes of novelty, I didn't find generating objects in the experience to be very interesting. The world of Wish Master is mostly just a big open space for people to generate things and run around in, so there wasn't much to do with the objects after I made them.

Roblox launches AI tech that generates functioning models with natural language

Roblox launched an artificial intelligence technology on Wednesday that can generate fully functioning in-game models using natural language prompts, as the company aggressively pursues machine learning to attract more developers to its videogame platform. The technology, dubbed 4D creation, released into beta, and builds upon the company's earlier model which could generate static three dimensional objects. The new model allows users to generate a vehicle, interact with its doors and drive it with accurate physics mechanics. The move serves to lower the entry barrier to Roblox's developer community, which is user-led and responsible for the games played on the platform, as the firm looks to spur the creation of new titles and boost its already sprawling user base. The company had more than 150 million average daily active users at the end of the third quarter. "You want to create an object, that is one path, and for the artist, the visuals might be easy, the coding might be hard. The other is, you're a game developer who finds the visuals harder, coding easier. So we're trying to bring all of them together, and our highest goal would be that a player can create inside a game," Roblox's senior vice president of engineering for foundation AI and infrastructure, Anupam Singh told Reuters. Roblox has been investing heavily in expanding server capacity to accommodate the sharp growth in its player base, while also putting funds towards developing AI models and safety - an issue that has come under heavy scrutiny from American state and international governments. 4D creation is a part of Roblox's larger push to develop AI world models, which can understand the rules and dynamics of an environment so that it can predict and generate future gameplay. The announcement comes a week after Alphabet's Google GOOGL.O launched an AI model allowing users to simulate and generate a real-world environment through prompts or uploaded images.