“AI will mean success for creators is based on the quality of their ideas”
Stefano Corazza, head of Roblox Studio, discusses how AI tools speed up the game creation process on its platform.
Welcome to AI Gamechangers, your weekly deep dive into how artificial intelligence is transforming the games industry. Each edition features a conversation with a leader working at the intersection of games and AI, bringing you practical insights from the frontlines of innovation.
Your reading for Yule week is this chat with Stefano Corazza, the head of Roblox Studio, conducted in conjunction with our pals at PocketGamer.biz. The online game creation platform Roblox needs little introduction. It’s a true phenomenon, at one point drawing in over half of all American children. Its engine is Roblox Studio, where creators build their own experiences. In 2024, new generative AI tools became part of Roblox Studio to help speed up the game creation process. We dive into what that means and Corazza’s philosophy for AI-enhanced creation in the Q&A below.
We have some fascinating Q&As planned for the months ahead, so please remember to subscribe. In the new year, you can look forward to conversations with ArenaX Labs about AI agents for multiplayer matchmaking and ii Connection about ethical AI recruitment.
Scroll to the end for some AI news to take us into the festive season, including the co-founder of Pixar joining the board of 3D world platform Odyssey Explorer, Planet Mojo’s poker-playing AI agents, a new AI startup from the former Twitch CEO, Databricks landing $10 billion in investment, fresh products from Nvidia and OpenAI, and loads more.
Stefano Corazza, Roblox Studio
Meet Stefano Corazza, who became head of Roblox Studio in 2022. In this conversation, we explore how Roblox is integrating generative AI tools across its platform to make development accessible to creators of all skill levels. Corazza presents an optimistic and strategic vision, balancing Roblox’s business ambitions with creator empowerment, emphasising how AI will enhance rather than replace human creativity.
Top takeaways from this conversation:
Roblox is developing "4D generative AI" - going beyond static 3D asset generation to create objects that incorporate multimodal understanding across appearance, physics, and scripts.
AI shows measurable impact on creator productivity. Roblox creators using Assistant and other tools increased what they could publish by 31%.
Language barriers in global communities are broken down by real-time translation. The Roblox communication system handles 20 billion messages and can even translate signs on in-game assets.
The economics of game development are shifting thanks to AI tools. The development cycle is accelerated by the automation of technical tasks, enabling faster iteration and market testing while preserving creative control. It’s particularly beneficial to smaller studios with limited resources.
AI Gamechangers: Both UGC and AI are hot topics. How do you see AI transforming user-generated content creation over the next few years?
Stefano Corazza: AI is already transforming creation. We believe that AI will mean that success for all creators is based on the originality and quality of their ideas rather than the ceiling of their technical skills. This mindset is all about lowering the barrier to entry and facilitating creation of anything for anyone, anywhere. We want to unlock the creativity of millions of creators to build better, more diverse UGC scenes, games, and items across Roblox.
Ultimately, we’re building towards what we call 4D generative AI, where AI is not just used to generate assets in 1D (like lines of code), 2D (like new textures for objects), or even 3D (entire objects, like avatars or scenery), but to generate 3D objects that can have dynamic interaction with users. This will require a model capable of multimodal understanding across appearance, shape, physics, and scripts, and we’re really excited to continue building it to support AI-powered UGC at a revolutionary scale, and we are able to build our dataset with explicit permission from and transparency with our creators.
Roblox’s CEO stated a goal of reaching a billion daily active users. How crucial is AI technology in scaling content creation to meet that level of demand?
We believe 10% of all global gaming revenue will flow through the Roblox ecosystem –and be distributed within our Roblox community. Helping creators build their business is a key part in getting us there, and AI will not only help the community to create more experiences, but also to make existing ones richer and more interactive.
As we release new generative AI tools – like Assistant, Code Assist, and our texture generator and avatar animation tech – we’ll streamline creators’ workflows, help them reach new audiences globally, and allow them to get from idea to reality, faster, allowing for more content relevant to more users. AI also helps us grow Roblox by making it easier than ever for users and creators to take part in its economy.
For example, our conversational AI Assistant for creators has recently exited beta testing, and we saw creators who use Assistant and/or our other AI tools at least once increased their publishes by an estimated 31%.
AI also plays a role in bringing users together and scaling access on Roblox. Thanks to our real-time translation technologies, we’re making something possible on Roblox that isn’t even possible in the physical world – enabling users to relate to and engage with people and content from around the world. Our AI translation technology is already growing the community by breaking down language barriers, helping more people to access everything Roblox has to offer with no geographic restriction. In the future, we envision every part of a user and social experience on Roblox to be fully translated into their native language.
Since joining Roblox from Adobe, you’ve overseen the launch of several AI tools. Which emerging AI technologies are you most excited about implementing next in Roblox Studio?
I’m very excited about evolving Assistant into a true platform for any new functionality people want to add to Roblox Studio or to any experience. Instead of implementing new UX paradigms and tools, you can simply provide Assistant a set of examples and that functionality will be live and integrated with existing creation and coding workflows. The rate at which people can add functionality to Assistant is much faster than coding any bespoke functionality from scratch. It is a leap forward in software evolution.
Sounds like the adoption rate of Code Assist has been significant. How do you see AI-assisted development evolving beyond code suggestions?
It has been so exciting to see how creators are using Code Assist, as well as how we’ve been able to finetune the model in tandem.
“AI on Roblox is all about helping creators make their idea a reality, faster. The faster a developer can put their idea out in the world, the faster they can test it and receive live feedback”
Stefano Corazza
Code Assist is a valuable tool for developers of any experience level, but it’s not the only AI-assisted tool for creation. Launched into beta in December 2023, Assistant is a generative AI suite that makes creation on Roblox more accessible by using plain-language prompts and empowers advanced creators to build richer, more engaging experiences, faster. Code Assist is one piece of this puzzle, alongside answering questions, generating materials and more. In fact, we just rolled out several new capabilities within Assistant:
DataModel Context: Roblox has expanded the amount of context available to Assistant in order to reduce the need for overly prescriptive prompts. Now, Assistant has the context about all of the elements that make up a user’s experience including items like the environment, scripts, and 3D objects. For example, if a creator has a 3D object that includes leaves, Assistant can understand that this object is a plant or a tree, and apply edits to it without having to explicitly name it as such.
Script iteration: Previously, if Assistant generated a script that a creator wasn’t satisfied with or wanted to modify, creators would have to undo or remove that script and re-prompt Assistant with a new prompt. With script iteration, Assistant can now work with creators to modify their scripts until they achieve what they want to create.
Multi-Script Insertion: Assistant works with creators to dynamically modify existing scripts as well as iterate on scripts it generates.
Creator Store Insertion: Assistant can now insert 3D models from Roblox’s Creator Store to help creators populate their scenes. Creator Store is where creators can buy and sell assets made by and for the Roblox community, including models, images, meshes, audio, fonts, videos, and plugins. This can be helpful for prototyping new concepts.
Ultimately, our vision for AI at Roblox is that it accelerates creators’ workflows, automates mundane tasks, and allows them to focus on their creativity.
We've been working to provide our creator community with a new set of cutting-edge generative AI tools for high-quality 3D generation, and we imagine a powerful future where it is possible to generate any 3D object or scene you can imagine. With the combination of AI and our endlessly creative creator community, the future will go beyond anything we have imagined today with an open-source solution that is not just foundational for us, but foundational for the industry.
At the Roblox Developer Conference a few months back, we announced our work to build a 3D Foundational Model, an open-source 3D AI generative model. Our foundational model will be built to most closely address the needs of 3D content creators of today and tomorrow, allowing them to build scenes that are truly dynamic, dimensional, and customisable.
Our goal is to allow creators to define their desired style with this model — whether photorealistic, anime, low-poly, or steampunk — and generate objects and scenes that fit that style. For example, developers would be able to simply greybox their experience, and let AI build a world around the map, iterate on it by giving AI feedback or further adjust and improve their creations via existing tooling.
How do you envision AI changing the economics of game development, particularly for smaller studios?
AI on Roblox is all about helping creators of all kinds – from one person creating for fun, to a small studio working on a few games, to a large studio working with IP holders and brands – make their idea a reality, faster. The faster a developer can put their idea out in the world, the faster they can test it and receive live feedback from Roblox’s 89.9M daily active users. We believe that this new speed of development will enable creators to access Roblox’s economy and fine-tune their monetisation more efficiently than ever before.
With all the new tools, how will Roblox balance making creation more accessible through AI while ensuring creators can still develop distinctive styles?
Our goal is that AI on Roblox can learn from individuals’ art styles and build within them, allowing specific developers to generate worlds that naturally fit their vision.
You can already use our texture generator within Assistant to generate textures within a designated art-style, and Assistant’s new MultiScript Insertion tool will support this to allow creators to batch-edit scenes through a specific prompt. The combinatorial complexity and variety of AI models also helps generate unique outputs that the creator can then take in further unique directions.
As someone who leads both traditional and AI-powered creative tools, how do you address concerns about AI potentially replacing human creativity? How do you ensure AI enhancement supports rather than diminishes genuine social interaction and creativity?
Generative AI will not replace creators on Roblox, but it will ultimately help creators deliver more diverse content – benefiting both our creator community and our users.
Our focus is on building tools that eliminate mundane tasks, like replicating the same edit on thousands of objects, so that creators can focus on what they love doing – building worlds. Everything that we do is designed to help our developer and creator community make their ideas come to life and succeed on the platform.
Roblox recently introduced AI-powered chat translations, processing nearly 20 billion messages. How do you see AI tools evolving to further break down barriers between global communities?
We see AI translation as crucial to breaking down barriers on Roblox and bringing the community closer together.
Our custom multilingual large language model supports live chat translation in 16 languages, meaning. For example, a user who only speaks Korean can seamlessly chat with one who only speaks English.
We’ve recently gone one step further to launch a real-time translation API that works in-game to translate all user-generated text, not just chats. For example, if a sign in a scene made by Korean speaker reads “식료품점”, for an English speaker it would automatically read “Grocery Store”.
Moving forwards, we expect to continuously improve the accuracy of these tools, updating our models with the latest slang and user feedback in all languages we support (which will give us the architecture to implement AI translation for any new languages). We are also exploring exciting new upgrades, like voice chat translation so that a user in France and a user in Indonesia, for example, would be able to have a fluent conversation, live, on Roblox. AI translations can turn Roblox into a real communication hub for people all over the world.
How does Roblox view the balance between quick, AI-generated content and deeper, more customised experiences?
We’re designing our AI to support highly customised creation with detailed layers of creator input to finesse everything from lighting conditions to textures and scenery. We want the content coming out of AI generations to be just as editable and expandable as any content creators can generate manually, so quick initial ideas from AI can be taken into more in-depth workflows and expanded upon into more nuanced experiences.
“We believe that AI will mean that success for all creators is based on the originality and quality of their ideas rather than the ceiling of their technical skills”
Stefano Corazza
Ultimately, we aim to empower people to create whatever they want. The Roblox userbase will decide which experiences are the most fun, engaging, and exciting, and creativity will be rewarded, as it always has.
Anything else to share?
As we continue to grow, we are investing in the tools and resources for developers and creators to build the next generation of experiences users can enjoy on the platform. And that’s where I’m focused – on building technology that will enable creation of anything, anywhere, by anyone, and AI is a great way to support that.
Further down the rabbit hole
The week’s must-read news stories, all in one place:
Odyssey, the generative world model company, announced that Ed Catmull, co-founder of Pixar, has joined its board. It also revealed Explorer, its first image-to-world model that transforms any picture into a detailed 3D environment.
Planet Mojo has unveiled its AI agent-led Agents Of Poker game. Following the release of its initial Emma AI (which uses GPT-4o), Planet Mojo has revealed details of its first playable experience based on its WorldWide Agents platform. You’ll be able to play against bots who have their own fine-tuned LLMs, run their own strategies and can place bets from their own wallets.
Emmett Shear, the former CEO of games streaming platform Twitch, is launching an AI startup called Stem AI. Andreessen Horowitz is an investor in the company.
If you haven’t tried the Audio Overview feature in Google’s NotebookLM, you must. And it’s getting even more interesting, as Google just released an update enabling users to talk with the AI podcast hosts.
Last week, Google released new state-of-the-art video and image generation tools in the form of Veo 2 and Imagen 3. Veo 2 creates videos with realistic motion and high-quality output, up to 4K. It can create videos that are over six times the duration of OpenAI’s Sora. Imagen 3 is Google’s top-quality text-to-image model.
Analytics company Databricks has raised $10 billion in its recent Series J funding round (surpassing OpenAI's $6.6bn raised in October). The money will go towards acquisitions, expansions, and the development of new AI products. Databricks provides its Data Intelligence Platform to games studios, including Minecraft, SEGA, Ubisoft and more.
GPU firm Nvidia has been busy. Last week, it revealed its “most affordable generative AI supercomputer,” the compact Jetson Orin Nano Super Developer Kit, which fits in the palm of your hand and costs around £200. The firm claims it delivers a 1.7x gain in generative AI inference performance. Meanwhile, Nvidia also unveiled Meshtron, a new model for generating meshes for 3D models (here’s its Github).
OpenAI has been running its “12 Days of OpenAI” online event before Christmas. It started with the release of its o1 reasoning model. It just ended the event by releasing its “next frontier model”, o3. There’s also o3-mini, a smaller and more cost-effective version. It represents a significant advancement in AI, although right now, OpenAI is inviting researchers to participate in public safety tests to ensure the models are deployed responsibly. You can watch the announcement and benchmarks here: