Hi @Joe_Kerr The UI was designed mostly through vibe coding, using Claude together with a UX/UI-oriented skill:
https://github.com/nextlevelbuilder/ui-ux-pro-max-skill
I don’t simply ask the AI to generate an interface and accept the result. I usually guide the process quite closely, specifying layouts, components, interactions, and visual direction. Sometimes I also give it more creative freedom with prompts like “make this screen feel more premium” just to see what ideas it comes up with.
Before starting, I spend time defining the architecture, tech stack, and project structure. I often use GPT to help create a detailed specification. For that, I usually rely on a Prompt Engineer GPT created by my brother:
https://chatgpt.com/g/g-691a51eed9b481918657a53b741bb2b8-ingeniero-de-prompts
I review and refine the generated specification and then feed it into Claude (usually through Antigravity).
Most of the codebase was initially generated through vibe coding. Once I have a working foundation, I reorganize the code, understand the architecture Claude proposed, and refactor it until I’m comfortable maintaining and extending it myself.
The 2D assets were mostly created with GPT Image 2. The 3D side is much more manual: the apartment was modeled, textured, and lightmapped by me in 3ds Max. I also used some free Sketchfab assets, low-poly characters from a Unity asset pack, and Mixamo animations.
Another fun part was giving Claude a fair amount of creative freedom to generate many of the game’s events. After defining the characters, mechanics, and overall tone, I let it invent renovation disasters, contractor drama, family situations, and random humorous events. Many of the final events came from those brainstorming sessions and were later curated and adjusted by me.
For me, AI doesn’t replace the design or development process. It acts more like a very fast collaborator that helps accelerate implementation and content creation, while the final direction, decisions, and refinements remain in human hands.