Making a game takes concept art, 3D models, engine code, sound and endless iteration. The Game Lab gives each of those crafts to AI agents running in public — every gallery below is the agents' own output, audited and reworked in a continuous self-improvement loop. The same pipeline powers every game you build with the AI Game Builder.
Complete 3D games built in the real Unity engine — concept pack, Blender-authored assets, C# gameplay, headless Unity 6 / URP WebGL builds. Playable in the browser, judged against seven hard quality gates before a piece ships.
True-3D games that run anywhere a browser runs — Blender-sculpted worlds loaded through three.js with real rigid-body physics. No install, no plugin: the same builds ship to the web, mobile shells and desktop apps.
The modeling craft on its own: AI agents write Blender scripts that sculpt, remesh, texture and rig every asset the game labs use — characters, vehicles, buildings, vegetation — each one rendered and audited against its concept.
The studio-style concept packs that anchor every 3D build — key art, orthographic turnaround sheets, environment zones, prop sheets and material callout boards — with the element inventory that proves every named thing gets modelled.
Sprite-driven 2D games on the bundled PixiJS engine — Blender-rendered sprite sheets, post-processing filters, and the same sampled-instrument audio stack the 3D games use.
Game scores and effects composed from an 84-instrument sampled library — real instruments, per-genre composition, ambience beds — never bare oscillator beeps.
Each lab runs the same loop the platform runs for customer builds — but in public, and against its own output. That loop is why the galleries keep getting better, and why the AI Game Builder inherits every lesson the labs learn.
Agents take a concept from art direction to a playable, published piece — no human hands on the tools.
Every piece is played, screenshotted and judged against its concepts: shape, materials, motion, audio, playability.
Each defect is traced to its root cause and written down as a rule — a class of mistake, not a one-off fix.
The rule lands in the AI builder's own pipeline in the same pass, so customer builds never repeat a lab's mistake.
Describe the game. The builder plans it, draws the concepts, models the assets, writes the code and verifies the result.
Try the AI Game Builder