At the Game Developers Conference on Monday, Roblox Corporation introduced a new set of AI tools aimed at “democratizing” game creation, allowing its millions of player-creators to construct functional game code and in-game 2D surfaces with simple text descriptions.
Head of Roblox Studio Stef Corazza highlighted the potential of the Roblox Code Assist beta, saying it can make game creation accessible to those “blocked by technical hurdles but had a great idea.”
With the new release, users can easily generate code snippets based on a set of defined objects, ultimately simplifying the game creation process.
As an example, Corazza demonstrated that a user can ask the system to “make orb turn red and destroy after 0.3 seconds when player touches it.” Based on the coder-defined orb object prompted in the code, seven lines of Lua function will be generated by the system.
Similarly, a prompt to generate a function that “creates a 3 by 3 grid of orbs around orb” will result in a few lines of code that will place a small grid of orb objects in the game scene.
Advancements in natural language code generation
Four months ago, it was uncertain whether the AI tool would work well enough to be released publicly. However, Roblox has leveraged recent advancements in natural language code generation to develop the tool further.
During the conference, Corazza said fine-tuning the standard model with code from the Roblox platform was crucial in getting usable results for the Code Generator Beta as it significantly improved the output quality.
Later, he also highlighted the need for context also applies to coders who use the tool. He compared asking the AI to generate code on a blank document to asking a knowledge expert to take a test in a completely white room where they didn’t hear the question correctly. The output quality will depend on the context in which the tool is used.
During internal testing, providing the AI tool with just three lines of sample code increased the “acceptance rate” for the tool’s suggestions by 50 percent compared to attempts that started with no such “context” code. Currently, the primary focus of the Code Generator Beta is to simplify the work of experienced coders by automating basic coding tasks.
In the future, Corazza envisions a chatbot-style interface that can act as a learning tool, elaborating how code works and documenting functions for those still learning the basics.
Roblox has also released a Material Generator tool that creates flat art assets layered on top of 2D surfaces in a game world.
Unlike basic image generation tools, this tool automatically layers a faux 3D normal map onto the surface alongside other maps for attributes like colors, roughness and metalness to enable accurate lighting reflections and responses to other objects.
However, the company’s AI asset-generation plans go beyond flat surfaces.
Roblox is developing an AI system that can create an entire specific geometry to re-texture a full 3D model or character entirely. This is a challenging problem because of the need to be aware of the object’s full context, but the team has seen “some early breakthroughs,” and the company is confident it will be available soon.
Corazza said the ultimate goal is to have an AI tool that can mimic a specific game style entirely.
Later, a user can type a prompt like “Scene with a forest, a river, and a large rock,” and get a fully interactive, realistic 3D world matching the prompt. The company estimates that this advanced tool will be publicly available in the next two years.