A video game engine built for everyday people
Everyone has a game in them. Harmony is how it gets out.
Describe the game you want out loud, type it, or build it by hand on a full editor — the same project, three ways in. MIA, your studio partner, turns what you say into scenes, characters, sound, and a shippable build. No engine jargon required.
Voice mode
Speak it
Talk to MIA the way you'd talk to a friend who happens to know how to make games. Hands‑free, conversational, good for the couch and the whiteboard.
Prompt mode
Prompt it
Type instructions and get results you can see. Precise, repeatable, easy to share — write a line, watch the scene change.
Classical mode
Build it by hand
A real editor underneath it all — scene tree, viewport, inspector, logic graphs. Nothing MIA does is hidden from you.
Movement I · The idea
Game engines are built for engineers. The people with the best ideas usually aren't.
A parent with a bedtime story worth playing. A teacher who wants an interactive lesson. Someone who's been describing the same game to friends for ten years. The tools exist, but the on‑ramp is a cliff. Harmony replaces the cliff with a conversation.
Who it's for
Idea people
Hobbyists, educators, tabletop designers, streamers, kids and the adults helping them — anyone who can describe a game clearly but has never opened an IDE.
The wedge
Talk, don't configure
The first ten minutes decide everything. In Harmony they end with a playable prototype, not a tutorial about coordinate systems.
The promise
No ceiling
When you outgrow the conversation, the classical editor is right there — same project, nothing locked, no export‑to‑a‑real‑tool step.
“Harmony is the conductor's assistant — you hear the whole piece, MIA handles the instruments.”
Movement II · Three ways in
One project. Voice, prompt, or the full editor — switch mid‑sentence.
Modes aren't skill tiers or separate apps. They're input methods over the same live project. Dictate a level on the couch, refine a value by typing, then open classical mode to nudge a collider by hand — all without losing your place.
Voice mode
Conversational
Wake word plus natural back‑and‑forth, like Siri or Alexa but scoped to your game. MIA reads results back, asks clarifying questions, and narrates what changed so you can keep your eyes on the screen — or off it.
- Push‑to‑talk or hands‑free
- Live transcript you can correct
- “Undo that” / “show me options”
Prompt mode
Typed & precise
A command box with history and a visible plan before anything executes. Best when you know exactly what you want, want to repeat it, or want to hand the recipe to someone else.
- Preview the plan, then approve
- Re‑run and tweak past prompts
- Prompts are saved with the project
Classical mode
The real editor
Scene hierarchy, 2D/3D viewport, inspector, asset browser, visual logic graphs, and an optional script view. Everything MIA generates lands here as normal, editable objects.
- Direct manipulation & snapping
- Node‑based logic, no code required
- Full undo history shared with MIA
Movement III · From idea to prototype
Design, create, refine — with a scene manager and asset manager holding it all together.
The notes call these Design mode, Game Creation mode and Prototype refinement. They're phases of one loop, not a wizard you finish once. The Scene Manager and Asset Manager are always open underneath, so nothing MIA makes ever goes missing.
Design mode — shape the concept
Talk through genre, mood, core loop, characters and win condition. MIA proposes a one‑page design doc and a matching starter template. Nothing is built yet; you're agreeing on the target.
Game creation mode — build the first playable
MIA assembles scenes, blocks out levels, wires the core mechanic, drops in placeholder art and sound, and hands you something you can play inside a minute or two.
Prototype refinement — make it good
Iterate on feel: tune jump height, swap the palette, add an enemy type, replace placeholder assets with generated or uploaded ones. Small asks, fast turnarounds, always playable.
Scene Manager — the map of your game
Every screen, level and menu as a board you can reorder, duplicate and link. Set the entry scene, define transitions, see which scenes share assets.
Asset Manager — one shelf for everything
Art, audio, models, fonts and templates in a searchable library. Tracks where each asset is used, its licence and origin (uploaded, generated, or from a template), and keeps versions.
Movement IV · The suites
Three focused workspaces for the parts that deserve their own room.
Most work happens in the main modes. But design, sound and marketing each reward dedicated tools, so Harmony gives each one a suite — reachable by voice (“open the sound suite”) or from the editor.
Suite 01
Game Design
The tuning bench. Balance stats and economies, edit levels on a grid, manage the visual logic graphs, run playtest sessions with recorded input, and A/B two versions of a mechanic side by side.
- Level & encounter editor
- Balance tables with live preview
- Playtest capture & heatmaps
Suite 02
Sound
Score, effects and voice in one mixer. Generate or import music, place adaptive layers that respond to gameplay, design SFX from prompts, add ducking and buses, and preview against live play.
- Adaptive music layers
- Prompt‑to‑SFX + sample library
- Mixer, buses, ducking, mastering
Suite 03
Cinema
Trailers made easy. MIA cuts a first trailer from your actual gameplay, then you refine it on a simple timeline — beats, text cards, music sync, capture in‑engine at any resolution, export for every store.
- Auto‑cut from real playthroughs
- Beat‑matched music & titles
- Store‑ready presets (PC, console, mobile, social)
Movement V · Meet MIA
MIA is the layer between what you mean and what the engine does.
Like Siri, Alexa or Boardy — a named assistant you speak to plainly — but she lives inside your project and understands it. She holds the context of your whole game, plans multi‑step changes, and keeps every action reversible.
What MIA does
Understands the project
Reads your scenes, assets, mechanics and past decisions so “make the boss harder” means the right thing. Asks before guessing when a request is ambiguous.
- Plans, then acts — shows the steps for anything non‑trivial and waits for a nod
- Routes generation — sends art, audio, models, music and logic to the right specialist model
- Narrates changes — a plain‑language changelog you can read or hear
- Teaches gently — names the classical‑mode control it just used, so you learn the tool by using MIA
Where MIA stops
You stay the author
MIA proposes; you decide. Guardrails keep the assistant honest and the costs predictable.
- Reversible by default — every action is a labelled entry in shared undo history
- No silent spend — generation that costs credits is previewed and confirmed
- Nothing hidden — every MIA edit is a real object you can open and change in classical mode
- Provenance tracked — generated assets are tagged with model and prompt for licensing clarity
Voice mode. Plain speech in, a confirmed plan and a spoken changelog out. The same exchange typed is prompt mode.
Movement VI · Under the hood
A conventional game engine with an orchestration layer sitting on top.
The bet: don't reinvent the runtime. Build (or license) a solid ECS engine, then invest everything in the MIA layer that translates intent into engine operations, and in the pipeline that feeds it content.
Project on disk
game.harmony # manifest: title, targets, entry scene
scenes/
title.scene level-01.scene win.scene
assets/
art/ audio/ models/ fonts/
logic/
player.flow # visual graph
spawn.script # optional text
mia/
context.json # what MIA knows
history/ # prompts + actions, all undoable
build/ # output per target
How one request flows
Capture
Voice → text, or typed prompt. Attached to the current scene & selection.
Plan
Intent parser + project context produce an ordered list of engine ops and generation jobs.
Confirm
Plan shown to you. Anything that spends credits or deletes work needs an explicit yes.
Apply
Ops run against the runtime; generated assets land in the Asset Manager; one grouped undo entry is written.
Report
Plain‑language changelog, spoken and written. Preview updates live.
Movement VII · The interface
Schematic mockups — the shape of the tool, not the final pixels.
Two more surfaces beyond the voice exchange above: classical mode, where every generated object is directly editable, and the Cinema suite timeline.
Scene tree
Inspector — Player
Logic
Classical mode. The lantern MIA created in voice mode, opened as normal objects — sprite, light, and a fuel value wired in a visual graph. Change any of it directly.
Cinema suite. MIA's first cut from real playthroughs, on a four‑track timeline. Titles snap to music beats; export presets cover every store.
Movement VIII · Publish, templates & accounts
Start from a template, save to your account, ship to any screen.
Prototype templates
Starting points
A curated set MIA picks from during Design mode: platformer, top‑down adventure, endless runner, puzzle‑match, visual novel, tower defense, first‑person walk‑and‑talk, card battler. Each is a working game, not an empty scene.
User logins
Accounts & projects
Email or social sign‑in, cloud‑saved projects with version history, device sync, and a generation‑credits balance. Family‑friendly defaults; a mode for classrooms with shared project libraries.
Ownership
Your game is yours
Export the full project at any time. Generated assets carry provenance tags. Clear terms on what you can sell and where.
| Target | How it ships | Notes |
|---|---|---|
| Windows / macOS / Linux | One‑click signed build from the cloud build farm | Available first — no third‑party approval needed |
| Web | Hosted playable link | Great for playtests, jams and sharing |
| iOS / Android | Store‑ready package + guided submission checklist | Needs the user's developer account; Harmony handles the build |
| Consoles | Build pipeline via platform partnerships | Requires platform approval & certification — later phase, dev‑kit gated |
Movement IX · Build order
What ships first, and what waits.
The riskiest assumption is that MIA can reliably turn plain speech into good engine operations. Everything in the MVP exists to test that as fast as possible on one genre.
MVP — prove the loop
In scope
- Prompt mode (typed) + classical editor
- MIA planner over one template genre (2D platformer)
- Scene Manager + Asset Manager
- Placeholder art/audio generation
- Accounts, cloud save, PC + Web export
- Shared undo history
V1 — make it broad
- Voice mode + spoken changelog
- 3–4 more template genres
- Sound suite
- Higher‑quality asset generation + upload
- Mobile export
- Collaboration (shared projects)
Later — make it a platform
- Cinema suite
- Full Game Design suite (playtest capture, A/B)
- 3D templates
- Console pipeline via partners
- Marketplace for community templates & assets
- Classroom / education tier
Phase 0 · internal
Engine runtime chosen (build vs license). MIA planner prototype driving the editor for a handful of scripted commands on the platformer template. Goal: does the intent‑to‑ops translation feel magic or frustrating?
Phase 1 · closed alpha
MVP scope in the hands of ~50 non‑developers. Instrument everything: where MIA misunderstands, where people drop to classical mode, where they get stuck. Goal: a first‑session that reliably ends in a playable prototype.
Phase 2 · open beta
Voice mode, Sound suite, more genres, mobile export. Public sign‑ups, generation credits live. Goal: retention past week one and the first games published by strangers.
Phase 3 · 1.0
Cinema suite, collaboration, education tier, marketplace. Goal: a creator can go idea → prototype → trailer → PC/mobile release entirely inside Harmony.
Phase 4 · consoles
Platform partnerships and certification pipeline. Goal: one‑click console builds for eligible projects.
Coda · Open questions
The honest list — what could sink this, and the current thinking.
AI art and audio that don't match each other break the illusion of a real game. Style drift across a project is worse than any single weak asset.
→ Per‑project style locks; a house art direction per template; regenerate‑to‑match rather than regenerate‑fresh.
“Everyday people” can hit a wall the moment they need real logic. If dropping to classical mode feels like being handed Unity, we've just moved the cliff.
→ MIA always names the control it used; visual logic only, no code required for the MVP genre; refinement asks stay small.
Voice transcription plus multi‑model generation on every iteration can cost more than users will pay.
→ Credits model with a generous free tier for placeholders; cache and reuse aggressively; cheap local models for planning, premium models only for final assets.
Platform holders gate SDKs, certification and store access. “Publish to consoles” depends on partnerships that take years.
→ Positioned as a later phase; PC, web and mobile carry the value proposition on their own.
Building a runtime is a multi‑year effort that isn't the differentiator. Licensing one constrains the product and the business model.
→ Evaluate an embeddable open‑source runtime for Phase 0; treat the engine as replaceable behind the MIA layer.
Spoken instructions are vague (“make it pop more”). Misfires that silently change the game erode trust fast.
→ Plan‑then‑confirm for anything non‑trivial; tight, reversible steps; “undo that” always one phrase away.