This is a prototype, not on any store. There is no link to give out, and the name is still a placeholder — "Defend." The owner tries it first, then decides what comes next.
What the owner pictured was flashy: four skills per hero, summons, monsters swarming in so an area attack can wipe them all at once, and a boss that shows up on a condition rather than a fixed round. The idea was to remake the old "defend the base from four sides" wave-defense games.
The first thing to decide — what to build it with
Say "let's build a game" and Unity is the first thing that comes up. But every app running today uses the web approach — built with web tech, wrapped in Capacitor for the app store. Whether to switch had to be settled first.
What the owner said
Make a game like the old ones — several heroes using skills to fight off swarms of monsters
So here is what I did
- Didn't start coding right away — checked whether switching to Unity made sense first
- Pulled subtitles for 87 of 364 videos from Unity Korea's official channel — 2D, mobile optimization, ads/billing, AI tooling
- Separately checked whether the art could be AI-generated (below)
- Conclusion: not switching to Unity for now. Built the stage-1 prototype in Phaser, a game engine that runs in the browser
What the owner diddesigned the direction (characters, skills, boss rules) and tried it on a real phone
Watched 87 official Unity videos
Before deciding, the subtitles of Unity Korea's official videos got read start to finish. A lot of it applies no matter which engine gets used.
- One ad network alone underperforms — the current standard is mediation, auctioning the impression across several networks
- A rewarded ad only pays out on the "watched to the end" signal — closing early gets nothing
- Plan against a ~10ms budget out of the 16ms for 60fps, not just an fps number
- When AI writes the code, writing gets cheap and verification gets expensive — so logic and screen stay separated, to keep automated tests easy
So Unity — not yet
| Criteria | Phaser (what's in use now) | Unity |
|---|---|---|
| Install / run | Opens straight in a browser, nothing to install | A separate engine to install and learn |
| Current apps | Several already run this way | Would mean starting from scratch |
| Where it wins | A reasonably-scoped 2D game | Real-time action, physics, a lot of enemies on screen at once |
The current scope doesn't need Unity. The note left for later: if Unity ever does make sense, start with Unity 6.3 LTS and set up the tool that lets AI drive the Unity editor directly before anything else.
The art wasn't AI-generated
The next question was how to keep drawing new, flashy characters. A few AI art services got a look.
- AI art services allow commercial use, but none had documentation for exporting transparent backgrounds or per-direction walk/attack animations in a form that drops straight into a game
- An AI layout tool (Google Stitch) was tried for character and object art too, but the style didn't match what was already in use, and some images came out broken
Character art won't be AI-generated.
If a tool has no documented way to export game sprites, testing it eats the time instead of saving it
The prototype itself
Stage 1 is a single, small map, not the wide one planned later. Monsters swarm in from every direction; a joystick moves the hero and attacks fire automatically. Anything outside sight range is covered in black.
| What was tested | Result | Fix |
|---|---|---|
| An idle bot auto-testing difficulty | Died on round 10, 233 kills | Raised the defended object's HP to 360 |
| The blueprint's 4-direction character art | The purchased pack only had a side-view sprite | Currently just flips left/right — reported to the owner as-is |
What's next
- The owner tries it on a phone and gives a read on the feel
- Stage 2 — leveling, skill unlocks, gear, saving progress
- Multiplayer was decided on from the start, but only after reporting the monthly server cost up front
- For the constant stream of new characters: the first few by a human, the rest possibly through an AI art subscription — no payment made on that yet
