How did you use DreamLayer?
Almost every picture in the game came from DreamLayer, about 72 of the 100 API credits: 15 overhead city maps, 200 characters, and the title, briefing and interface screens. The only exception is a few things generated in code (especially walking sprites, see below.)

Faces: one credit buys a sheet of about 72 front-facing portraits, so I generated several sheets and cut them with code (237 portraits including disguise looks). The game is about telling disguised people apart, so consistency mattered most.

DreamLayer did not hold identity across hats, glasses and hair dye, so I matched looks by hand, fixed skin tone and head size in code. Some faces were too similar for the gameplay (which relies on distinguishing them quickly), so I ran a DINOv2 face-similarity check to ensure accidental look-alikes never share a shift.

Maps: each city is one DreamLayer picture; a street recognizer I wrote turns it into the walkable street graph.

Screens: made from one shared style brief; a year or number in a prompt got painted into the picture, so the prompts avoid digits.

What failed: a top-down walking-sprite experiment (about 11 credits). A reference image fixed the camera angle and the sprite tool kept it across frames, but the figures had oversized heads, thin limbs and long strides that read badly at map size. People are drawn in code instead; the tiny size of the walking sprites (even at max zoom) meant generated assets were unnecessary for them anyway.

Before/after sheets (placeholder streets vs DreamLayer maps, placeholder vs DreamLayer faces, interface with and without art) are in the screenshots.

Game Engine / Framework
Godot