Summary
Highlights
Mark from the Studio Builder Tools team introduces procedural models, a new 3D space API. This feature allows creators to parameterize data models using Lua code to generate instances from parameters, offering greater control. The Assistant can also generate this code from prompts or images, enabling rapid creation of parameterizable content.
Mike Pleshi, a technical 3D artist, showcases the application of procedural models in building an 'Alchemist Lab.' He demonstrates how he used AI to generate initial inspiration images, extract objects, and then built various elements like the house, windows, and bookshelves. This highlights the power of procedural models in saving time and adding intricate details with customizable parameters.
Lynn, a principal product manager, introduces the planning mode for Roblox Assistant. This new feature completes the 'plan, build, test' agentic loop, providing a structured approach to game creation. Creators can access it through the Roblox Assistant or Studio MCP servers, benefiting from functionalities like procedural model generation for building and a playtest sub-agent for testing.
A quick demo illustrates the planning mode where creators can insert an image and ask the AI to generate a plan. The Assistant asks clarifying questions to align with the creator's intent, generates a task list, and then proceeds to build and test. Tips for using the planning mode include using it for complex tasks, reviewing and editing plans, and leveraging image inputs for better representation of ideas.
Peter McNeal presents the new scene analysis tools designed to provide direct insight into memory usage for Roblox projects. These tools offer views for audio, animation, script, and unparented instance memory, helping creators track down memory leaks and understand project composition. The instance composition view, for example, shows the number of instances of each type.
Peter demonstrates how the audio memory view identifies large memory consumers, and the animation memory view helps track animation loading and garbage collection. The script memory view offers a detailed analysis of Lua memory usage, making it easy to spot memory leaks in module scripts. The unparented instances view helps identify instances that exist but are not in the workspace, indicating potential leaks.
The scene triangles view provides an enhanced breakdown of draw calls and triangles, including terrain, grass, and particles, for render optimization. Furthermore, the video highlights the integration of AI with scene analysis. Using a 'scene health' command in Cloud Code, the system can automatically analyze a project, apply optimization knowledge, and provide assessments and guidance for improving performance and fixing issues like memory leaks.