Bloxig - AI That Makes Roblox UIs Playable

Roblox has millions of creators, and anyone who wants polished UI hits the same wall: designers mock UI in Figma, then developers rebuild every frame and write every button by hand in Studio. Existing tools convert how a UI looks, but the result is static buttons don't click, close buttons do nothing. A design file has no interaction logic, so no tool generates it. Bloxig solves both halves. It converts Figma designs into native Roblox UI faithfully, then its AI writes the Roblox code that makes them playable. Click a claim button and it responds; click the close button and the panel closes with the interaction code generated for you. The core innovation is structure-based detection. Instead of relying on layer names (which users write unpredictably "Btn_Dismiss", "Group 47", or in any language), Bloxig detects interactivity by structure, position, and text: a text-bearing box at button scale is a button; a small square in the top-right is a close button; an evenly-spaced row is a tab bar. Deterministic heuristics find the candidates, the AI infers each role and writes the Luau, and a verification pass guarantees every element is wired. This runs on AMD compute end to end. The interaction model Kimi K2, a top open coding model is served through the Fireworks AI API, which runs inference on AMD Instinct GPUs. Every generated LocalScript is written on AMD silicon. Bloxig is a real, deployed product a live web app, a published Figma plugin, and a Roblox Studio plugin not a prototype. The same approach used by commercial design-to-code tools, applied to Roblox, a platform none of them serve.