← Duckdoku home

🦆 How I built Duckdoku

A duck logic puzzle, built solo with AI tools

by Tetra Developers

Duckdoku is a small Sudoku-style puzzle game: one duck per row, one duck per column, one duck per color, and no two ducks touching. No numbers, no math — just looking and thinking. This page is the honest story of how it got made: which tools I used, what I paid for art, and what came from an asset store instead of being generated.

I copied the idea for Duckdoku from Meowdoku, a cat version of the same puzzle already on Google Play. I liked the rules, so I rebuilt the game myself with my own ducks, art, and sounds, in my own style — not a copy-paste of its code or art.

→ Read the full day-by-day development journal

Get it on Google Play Download on the App Store

The tools

I built the whole game inside Claude Code, Anthropic's coding assistant, talking to it like a teammate instead of writing every line myself. The model behind it was Claude Sonnet 5, on the $20/month Claude plan — no extra API bill for the coding side.

Claude didn't just write code in a text editor and hope it worked — it actually ran and drove the Unity Editor directly, using two things:

That combination is how features got built, tested with real gameplay (not guesses), screenshotted, and fixed in the same conversation.

On tokens: Claude Code doesn't hand me a running total of tokens spent across a project like this one, built over many separate conversations across several days — so I can't give an exact number here without making one up. What I can say honestly: it was a lot of back-and-forth, a lot of small fixes, and more than one rebuilt Android app bundle along the way.

The art

Image cost at a glance

Image modelGPT Image 2 (gpt-image-2), OpenAI image API
QualityMedium for all final art
Final images13 (7 sprite sheets, plus logo, win and lose sheets, app icon, splash screen, featured image), cut into 58 sprites
Total paid for images$0.85 on the OpenAI dashboard (26–27 September)

Breakdown of the 13 final images. Output price is real ($30 per million tokens). The reference-picture price is my estimate ($8 per million tokens), so each line is an estimate:

#ImageSizeOutput tokensEstimated cost
1Duck faces sheet1920×10881,325$0.041
2Dance and win sheet1920×10881,325$0.054
3Duck actions sheet1920×10881,325$0.053
4UI kit sheet1920×10881,325$0.042
5Streak and rewards sheet1920×10881,325$0.041
6Avatars sheet1920×10881,325$0.053
7Logo and splash sheet1920×10881,325$0.053
8Logo1536×640716$0.035
9Duck win animation sheet1920×10881,325$0.053
10Duck lose animation sheet1920×10881,325$0.053
11App icon1024×10241,756$0.066
12Splash screen1088×19201,325$0.061
13Featured image1536×768916$0.048
All 13$0.65

This matches the dashboard for 26 September ($0.65). The dashboard also shows about $0.20 on 27 September that is not in my own log, so I cannot break it down. The later low-quality test (see below) cost about $0.09 more by my estimate. Only image costs are shown here; the coding was covered by the flat $20/month plan.

For the duck characters, icons, and store art, I gave Claude my own OpenAI API key. Claude generated the images as sprite sheets — several poses or frames in one image, cheaper than generating each one separately — and then cut each sheet into the individual sprite images the game actually uses.

Contact sheet of generated duck sprite sheets, before being cut into individual sprites

The generated sprite sheets, before Claude cut them into individual images

Every image was made with OpenAI's image generation API. Before generating anything, Claude estimated the cost first and got my go-ahead. The dashboard total for image generation:

OpenAI usage dashboard showing total spend for image generation

OpenAI usage dashboard — total image generation spend

Would a cheaper setting have worked?

All the art above was made at medium quality. Afterwards I re-made the same 7 sprite sheets at low quality to see what I would have saved. Same model (gpt-image-2), same size (1920×1088), same prompts word for word — only the quality setting changed.

MediumLow
Output tokens per sheet1,325148
Output cost per sheet ($30 per million tokens)$0.040$0.004
Time per sheet27–37 s15–27 s
All 7 sheets, estimated total*$0.34$0.09

* Output prices are real. The price of input tokens (the reference picture) is my guess, so the totals are estimates.

Sprite sheets side by side: medium quality on the left, low quality on the right

Medium on the left, low on the right

I made each sheet only once and did not test high quality, so treat this as one honest test, not a rule. Full numbers and the exact settings are in Day 9 of the journal.

The sound

Sound effects and music came from two places:

What it looks like

Unity Editor open alongside a live gameplay screenshot

Unity Editor, mid-development — features were tested with real gameplay, not guesses

→ Read the full day-by-day development journal

Play it

Duckdoku is free to play on Android and iPhone/iPad. If you like quiet logic puzzles, give it a try — and if you don't like something, or spot a bug, I actually want to hear it. This is a small solo project, so every bit of feedback goes straight into the next update.

Get it on Google Play Download on the App Store

Got a suggestion, a bug, or just want to say hi? — tetraheaddevelopers@gmail.com