A developer is testing a workflow to create a third-person game with AA-level graphics using AI tools, avoiding manual coding and asset creation. The project utilizes the Unity engine, MeshyAI for model generation, and Blender with the Riggify add-on for rigging. Animations are sourced from Mixamo, while the Cursor IDE serves as the primary coding assistant.
Initial experiments revealed significant challenges in AI spatial awareness. When tasked with building a game scene, the AI struggled to arrange buildings and objects logically, often resulting in overlapping structures and inconsistent scaling. The developer attempted to address this by importing detailed geospatial data from OpenStreetMap via a GeoJSON file, which helped define the layout of streets and buildings.
Further technical hurdles included compatibility issues between Unity 6 and older assets, as well as degraded PBR textures when using AI-based upscaling tools like Nano Banana. To manage object placement, the developer implemented a custom C# script that calculates coordinates relative to neighboring objects, ensuring proper spacing and scale within the game environment.
The current development pipeline focuses on refining the technical infrastructure before addressing gameplay mechanics. The developer plans to test additional tools, including Codex and custom-built harnesses, to improve the automation of the level design process and overcome the limitations of current AI models in understanding 3D space.
