When to use it
You need visuals for a game, and it matters that every sprite is in one style rather than a grab bag. This prompt doesn't produce a picture — it produces a strict style brief plus ready-made text descriptions for an image generator or an artist. The AI plays art director for a pixel game. The result: one set of rules that keeps every asset on the grid.
The prompt (copy and paste)
You are the art director of a pixel-art game. Build a single ART BRIEF that keeps every asset in one style.
GAME AND MOOD: "<FILL IN, e.g. cosy farming roguelike, warm evening>"
STYLE REFERENCES: <e.g. Stardew Valley, Dead Cells>
ASSETS NEEDED: <e.g. hero + 3 enemies, forest tileset, 5 items, UI icons>
Return:
1. The style rules: sprite size (e.g. 32x32), palette (8-16 colours with HEX and the role of each), outline thickness, light source and shadow direction, level of dithering.
2. For EACH asset — a copy-and-paste text prompt for an image generator: subject, pose/angle, key details, what NOT to add.
3. Animations: for the hero and enemies — the list of frames (idle, walk, action) and how many frames each needs.
4. Tileset: which tiles are needed for seamless tiling (centre, edges, corners, transitions) and the rules for how they join.
5. Technical requirements: format (PNG, transparency), grid, what goes into a single atlas.
6. A consistency checklist: how to verify a new asset is "in style" (palette, size, light, outline).
Specify the palette explicitly (HEX) — it's shared across all the prompts. Mark anything you filled in yourself as [assumption].
Why it works: hoisting palette, size and light into a shared rule set above the individual prompts is what keeps a set of assets coherent; without it the generator drifts off style every time.
Filled-in example
Game: "a dungeon roguelike, dark but readable". References: Enter the Gungeon. Assets: hero + 3 enemies, stone dungeon tileset, 4 item icons.
What the AI should return: size 24x24, a 12-colour palette with HEX (stone base, neon accent for hazards), a 1px dark outline, light from the top left; for the hero, a prompt like "pixel-art knight, 24x24, idle facing camera, 12-color palette [...], dark 1px outline, no background"; a walk cycle of 4 frames; a tile set of centre / 4 edges / 4 corners; a checklist of "12 colours, 24px, light from top left, outline present".
Variations
- Not pixel art. Swap "pixel" for "flat vector" or "hand-drawn" — the brief's structure is the same, only the style rules change.
- One asset, many states. "Give me prompts for the hero in 4 colour variants for different classes, shared palette."
- For a specific generator. "Rewrite the prompts for Midjourney with the flags
--ar 1:1 --tile" (or whichever tool you use).
Pro tips
- A HEX palette in every prompt is the cheapest way to hold the look together; without it the model drifts from asset to asset.
- The "what NOT to add" field (background, stray highlights, text) saves hours of cleaning up generated output.
- For pixel art, generators routinely lie about the grid: you'll still finish the pixel-level cleanup in Aseprite or another editor — the AI gives you a blank to work from.