In a one-person studio, every asset has the same owner: the person who also writes the code, answers support email and submits the build. So the question for any asset isn't only "does it look right?" but "how do I regenerate it when something changes?" An image exported by hand once is easy. Exporting it again in three sizes and two appearances after a palette tweak, six months later, is where time disappears.
The approach across our apps has been to make as many assets as possible either generated or captured by a script, and to keep the few hand-made ones in vector form. Here's what that looks like in practice.
Sound you don't have to store
The native version of Sudoku: Color ships with no audio files. Each sound — a tap, a
correct placement, an error buzz, a win flourish — is synthesised when the app starts,
using AVAudioEngine and a handful of functions that generate sine tones,
sawtooth tones, frequency sweeps and short arpeggios with a decay envelope. Each game
event maps to a buffer built once at launch.
The trade-offs are real. You won't get rich recorded sound this way, and tuning a tone means changing numbers in code rather than dragging a file in. In exchange, there's nothing to license, nothing to re-export, the app bundle stays small, and a sound change is a reviewable one-line diff. The audio session is set to ambient, so the effects mix with whatever the player is listening to and respect the silent switch.
Colors as code, not as images
All of the game's colors live in two Swift files. Theme tokens cover the interface — surfaces, text, borders — and switch between light and dark. Palettes hold the nine gameplay colors and don't switch. Because every screen reads from those tokens, a palette or theme change is an edit to one file, and there are no colored image assets to fall out of sync.
One vector source for the brand mark
The splash logo in Sudoku: Color is a single vector PDF traced from an SVG source kept in the repository. The system launch screen and the SwiftUI splash view both draw that same asset, at the same size, on the same background color (with light and dark variants). Because both stages use the identical image, the hand-off from launch screen to app is invisible.
The same rule applies to this website: the fox mark and the browser icon come from one traced vector, and the small PNG icons are rendered from it rather than drawn separately.
Screenshots from a script
App Store screenshots are the asset most likely to go stale, because they show the actual interface. For Sudoku: Color, a small build-runner script builds the app, installs it on a named simulator, and captures a sequence of screenshots in both light and dark appearance. The store screenshots kept in the repository are at the 6.5-inch iPhone size (1284×2778) and the 12.9-inch iPad size (2048×2732).
The script watches a queue folder and runs only a fixed list of jobs, so it can be left running without becoming a general-purpose remote shell. That last detail matters more than it sounds: automation that can do anything is automation you eventually stop trusting.
Gameplay capture for ColorPair
ColorPair goes a step further, because short video is a different problem from stills. The app has a demo mode that plays itself, triggered by a value written into the simulator's defaults before launch — more dependable than trying to deliver a URL to a freshly launched simulator app. A capture script records a stream of frames while the demo runs.
A Python pipeline then takes over:
- A moment detector scans the frames for visually interesting stretches — a board clearing, a puzzle completing, a burst of fast matches — and scores them.
- A renderer turns the best moments into vertical 1080×1920 H.264 video drafts, using Pillow for frame work and ffmpeg for encoding.
- A brief generator writes the candidate clips and suggested captions to a spreadsheet for review.
The output is drafts, not finished marketing. The value is that recapturing after a UI change is a re-run rather than an afternoon of screen recording and trimming.
What we'd suggest to another small team
- Ask of every asset: what regenerates this? If the answer is "me, by hand", decide whether that's acceptable before you have twenty of them.
- Keep brand marks in vector form and render raster sizes from them.
- Put colors in code as named tokens. Images of colors are the first thing to drift.
- Script screenshots early. The best time to automate them is before the first store submission, not the fifth.
- Keep automation narrow. A script that does five known things is one you'll keep using.