Index
A technique you can reuse, written out step by step.
14 pagesClear filters
Write a one-page spec (core loop, controls, win and lose, camera, scope limits, acceptance tests) before you prompt, then build in small milestones you verify by playing. What goes in the spec, how big a step should be, and when to restart.
How to make the 30–90 s trailer or gameplay demo that most of the AI-made games we studied were posted with: clean 60 fps capture, in-engine camera paths, a hook in 2 s, a cut on the beat, captions, and 16:9 plus 9:16 masters.
How to brief AI music, SFX and ambience, build seamless game loops and stingers, sync sound to cuts and gameplay events, hit platform loudness targets such as −14 LUFS, and avoid licensing traps.
How to let a coding agent drive Blender through an MCP server: setup, building rules and references, self-checks with fixed-camera renders, splitting work between agents, and keeping the scene editable.
The export recipe for getting Blender scenes into browser games as small, fast GLB files: phone and desktop budgets for triangles and draw calls, baking, texture sizes, Draco/Meshopt/KTX2 trade-offs, naming, colliders and LODs.
Shot sizes, camera moves and lenses in the exact words video models follow, when each move works, and how to build the same move with a Unity, Unreal, Godot or Three.js camera.
How to keep one character recognisably the same across 2D images, a reused 3D hero model and AI video clips: a master reference sheet, a fixed identity block, the reference features of current tools, and a drift check on every output.
Turn playtest words like floaty, slippery or weak into measured numbers (acceleration and stop times, coyote time, input buffers, camera damping, hit-stop, screen shake) and tune them with an AI model one or two values at a time.
The full path from one concept image to a game-ready 3D model with Tripo, Meshy, Rodin or Hunyuan3D: input image rules, picking candidates, decimation and retopology budgets, UV and texture fixes, scale, pivot and export.
Build AI video from designed stills instead of text: start and end frames, 5–10 s clips chained into a sequence with identity and light held steady, the best takes picked, then upscaled and interpolated.
How to split game work between separate AI agents (planner, builders, playtester, reviewer), connect them with task cards, hand-off notes and screenshot evidence, stop them undoing each other, and keep token costs in check.
How to turn photos of a real place into a playable level, with a desk race track as the example: capture and scale reference, blockout, modelled, generated or scanned props, collisions and the racing line, and lighting that keeps the real look.
The release checklist between a playable prototype and a link anyone can open: WebGPU-to-WebGL fallback, a loader that paints in under a second, progressive loading, touch and performance on phones, hosting, link previews and publishing on OpenDrama.
The seven layers that take a Three.js scene from flat tech demo to a console-grade image: colour management, tone mapping, HDRI lighting, one shadowed key light, fog, bloom and lens, with the frame-time cost of each.