I created the game and directed its art, and built v1 myself. For v3 I set the style, supplied the concept references and made every keep-or-cut call and the key gameplay decisions, while Claude Code wrote the code and the code-drawn pixel art from my feedback.
Why v3 dropped the AI sprite sheets
v2 swapped the characters for AI-generated sprite sheets. The checkerboard "transparency" was baked into the pixels, so every frame had to be cut out by script, and several late commits went into fixing mis-cut frames and stray white boxes. The sheets were also more detailed than the hero, so the two clashed on screen. v3 draws everything in code, and all 26 creatures share one set of outline and three-tone shading rules.
The hero: new clothes, same face
To bring the hero in line with the new monsters, six options (A–F) were rendered side by side: outlines, shading, bigger eyes, a full chibi redraw. My verdict was that the original face was cuter. The v1 face and ears stayed pixel-for-pixel, only the soft outline and shading carried over, the robes were slimmed down, and each realm got its own outfit design instead of a palette swap.
Squeezing concept art into 50 pixels
I supplied a detailed bestiary sheet. Used as-is, each beast would have been three or four times the mouse's size and would have needed image files. Instead each creature was boiled down to its signature features and redrawn at 40–60 px. One dragon went first as a test, and only after I approved it were the rest redrawn. A vertical S-curve dragon was also tried, but I found it less cute and it was reverted on the spot.
Pacing: breakthrough, bottleneck, tribulation
A headless simulator showed the old build reached Mahayana in about two hours with zero deaths, and the demon tower could be climbed past 5,000 floors. v3 works monsters out backwards from how many hits a reference player needs to kill them: 1.8 right after a breakthrough and 4.5 at the bottleneck, so a stall sends you back to upgrading. At 1× speed, Mahayana now takes about 3.4 hours.
An idle game doesn't need tension
The simulated player never died, so Claude proposed adding combat tension. I said no: players shouldn't have to think that hard, because idling alone should make you stronger. I kept only the realm boss, which became a reward rather than a wall. If you lose, it comes back 90 seconds later, and it never blocks your tribulation.
Player discussion
Loading discussion…