You hired a contract artist for eight tavern props. You sent them a board with forty images: firelit interiors, some Skyrim screenshots, a couple of oil paintings, three Artstation pieces you liked. Two weeks later the props arrive. The wood grain is photoreal, the silhouettes are busy, the palette is thirty degrees warmer than your existing kit, and nothing sits on your 32-pixel grid. Nobody lied and nobody was lazy. The board never actually said anything — it just pointed at things.
A mood board is not a collection. It is a set of constraints with pictures attached. If someone can look at your board and produce work you reject, the board was decoration.
The failure mode: collecting instead of constraining
Most boards are built by dragging in everything that felt right. That produces a document with a high variance of styles and no stated rules, which means the artist has to guess which axis you cared about. Did you pin that screenshot for the lighting, the silhouette language, the colour ramp, or the mood? All four are visible in the image. Only one of them was your point.
The fix is boring and it works: every reference gets a reason, every board gets a rule, and every rule gets a test you can run on a finished asset.
Write three art pillars, each with a test
Art pillars are the same idea as design pillars: short, opinionated, and capable of rejecting work. "Stylised fantasy" rejects nothing. Write pillars that can fail an asset.
Bad pillar: Cosy and hand-painted.
Better pillar: Readable at 25% zoom — every prop must be identifiable by silhouette alone in a black fill at 128 px wide.
Three examples that carry a test:
- Flat colour, no gradients in albedo. Test: open the texture, count unique colours per material; more than five means it fails.
- Light comes from one source per scene. Test: any asset with baked directional shading from a second angle gets rejected.
- Every interactive object is warmer than its background. Test: sample the hue of the prop and the wall behind it; interactables sit in the 20–50° band, environment sits blue-grey.
Three pillars is usually the ceiling. Five pillars is a wish list, and wish lists don't reject anything.
Annotate every reference in one line
The rule: no image enters the board without a caption naming what to take and what to ignore. One line each.
Tavern interior reference. Take: the way the ceiling beams cut the frame into thirds. Ignore: the photoreal soot and the warm bloom — we do not use bloom.
Character portrait reference. Take: shoulder-to-head ratio, roughly 3:1, reads as "heavy" at thumbnail size. Ignore: the rendering; we are flat-shaded.
This turns forty ambiguous images into forty small instructions. It also exposes duplicates — if six images share a caption, you only need one of them. Most boards shrink by half the first time you do this, and the half that survives is the half that was doing work.
Keeping the references and their captions inside the mood board module in GameDesignerX rather than a shared drive folder matters for one practical reason: you can link a board directly to the location, character or item entry it governs, so the artist opening the "Harbour District" location page sees the board that applies to it instead of hunting for the right folder.
The do / don't pair beats a wall of "do"
Pick your five most-likely-to-be-misread traits and give each one a matching negative example. A single contrasting pair communicates more than ten aspirational pins.
| Trait | Do | Don't |
|---|---|---|
| Outlines | 1 px, dark variant of the fill colour | Uniform black outlines |
| Proportions | 6-head heroic, hands oversized ~15% | 8-head realistic |
| Palette | 32-colour shared ramp, no off-palette pixels | Per-asset custom colours |
| Detail density | Detail clusters at focal point, quiet edges | Even detail across the whole texture |
| Materials | Metal reads by shape and highlight placement | Metal reads by specular map |
The "don't" column is where you put the mistakes you've already received. Every rejected delivery should add a row.
Lock the numbers before the first asset
Style is subjective; specs are not. A board that carries specs prevents the expensive class of rework — the kind where the art is beautiful and unusable.
The minimum set for a 2D project: canvas sizes per asset class (character 64×64, prop 32×32, tile 32×32), pixel grid, whether rotation is allowed, palette file with hex values, outline width, animation frame counts and frame rate, pivot point convention, and file naming.
For 3D: texel density (e.g. 512 px per metre), poly budget per class, UV layout rules, material count ceiling, up axis, scale unit, rig naming, and the LOD tiers you expect.
Put these at the top of the board, above the pretty images. Artists read the top.
Three tests you can run on any delivery
- Silhouette test. Fill the asset black on white. Can you name it? Can you tell it apart from the other five assets in its class?
- Thumbnail test. Scale to 64 px. Does the focal point survive? Most detail-density problems are visible only here.
- Neighbour test. Drop it into a screenshot beside three existing assets. Style drift is invisible in isolation and obvious in context. This is the one teams skip, and it is the one that catches the tavern props.
Run all three before you pay an invoice, not after you've built a level around the asset.
Keep the board alive
A mood board written once at the start of the project and never touched is a historical document by month three. Two habits keep it current:
- Promote shipped assets into the board. Once an asset is approved, it becomes a reference — it is more authoritative than anything you pinned from the internet, because it is actually in your game.
- Date and version it. When a pillar changes, note the date and what changed. An artist who started in March needs to know that the outline rule changed in June, and a one-line changelog on the board is cheaper than a meeting.
If your mood boards live alongside your lore bible and location entries, this happens by itself: the board for "Harbour District" gets updated when the district's description does, and the person doing the art sees both.
The checklist
Before you send a board to anyone — contractor, teammate, or your future self:
- Three pillars, each with a test that can reject an asset
- Every reference has a one-line take/ignore caption
- A do/don't table covering the five most misreadable traits
- Technical specs at the top: sizes, budgets, palette, naming, pivots
- Palette file attached with actual hex values, not a screenshot of colours
- At least three approved in-game assets shown as the current standard
- Named owner and a last-updated date
- One example of previously rejected work with the reason
Why this pays for itself
The board is doing three jobs at once. It briefs new collaborators without a call. It gives you language for rejecting work that isn't "I don't know, it just feels off" — which is the sentence that ends freelance relationships. And it forces you to decide your style before you commission forty assets in it, which is the decision most teams defer until they've paid for the wrong forty.
Spend an afternoon. Write the pillars, caption the references, fill the don't column with the mistakes you already know you'll make. The tavern props will come back right the first time.