What Four Public Game Art Bibles Actually Say

Almost every article about art bibles is written by someone who has not read one. They list headings — palette, silhouette, lighting — and stop there, because the headings are the easy part.

Four teams have published theirs in full, with the numbers still in. Reading them side by side is more useful than any amount of advice about what an art bible should contain, because it turns out they independently converged on the same five things.

They also share one blind spot, and it is worth naming before you read a word of them: not one of the four says anything about art that is generated rather than drawn. Every rule in all four assumes a human is making each asset and can be given feedback. For a lot of teams — probably including yours — that assumption stopped being true a while ago, and closing that gap is why we keep publishing this material. Read the four with it in mind. It changes which parts are worth copying.

The short version. The four public art bibles worth your time are 0 A.D.'s Art Design Document, the Liberated Pixel Cup style guide, Battle for Wesnoth's Creating Unit Art, and Riot's League of Legends VFX style guide. Every one of them specifies a tie-breaker, a light direction with the ambiguity spelled out, a shadow color as an actual value, a level-of-detail floor tied to a real-world object, and a named list of the mistakes people keep making. Every one of them also stops at the same place: they are built to keep two artists from disagreeing, not to keep one generator from drifting away from itself over forty assets. That second document is the one Makko exists to write.

The one thing none of them covers

Before the teardowns, the caveat that determines how you should read them.

All four documents were written to prevent one specific failure: two artists disagreeing. That is what an art bible has historically been for. Several people, several hands, several opinions, and a document that makes their output look like it came from one place. Every rule in all four is shaped by that problem — a shadow hex so two people pick the same shadow, a light direction so two people light the same way, a canvas size so two people hand in the same rectangle.

A generated pipeline fails differently, and much more quietly.

There is only one source, so it never disagrees with itself the way two people do. Every output is internally consistent — deliberate, confident, finished. The drift happens between generations. Asset forty has no memory of asset one. The armor trim creeps warmer. The outline thins. The light wanders around to the other shoulder. Nothing looks wrong on its own, which is exactly why nobody catches it until forty assets stand in one scene and the roster stops reading as one roster.

The document that prevents that drift is a different document, and the six items above are its table of contents. It is less concerned with what the art should look like than with what must not move while it is being made.

That is the gap Makko is built to close. The constraint layer that used to live in one artist's head has to be written down and enforced on every render, or generated art drifts — and it drifts in the worst possible way, looking fine asset by asset and wrong in aggregate.

What follows is still the best available starting point for the half these four documents do cover, and that half is real. Read them for it.

0 A.D. — Art Design Document

Wildfire Games. 21 sections, covering an RTS with thousands of units on screen.

The most complete one anybody has published, and the one most worth stealing structure from. It runs from artistic theme through concept art, models, skins and textures, animations, folder structure and file naming conventions — the full production span rather than just the look.

Two things it does that almost nothing else does.

It states a tie-breaker. Eight words: "Gameplay trumps Realism when the two topics disagree." The worked example is phalanx formations, which historically ran to 500-plus men and are capped at roughly 20 units because of screen space and pathfinding. That is a document that has already had the argument and written down who won.

It scales budgets by how much the player looks at something. Texture sizes are assigned by gameplay importance rather than by object type:

Both rules travel. The ladder survives translation to 2D as canvas sizes rather than texture sizes, and the detail floor survives it untouched — a human hand is a human hand at any resolution, which is why that one sentence has outlived the engine it was written for.

Liberated Pixel Cup — style guide

OpenGameArt, with Mozilla and the Free Software Foundation. Written so strangers could contribute art that matched.

This one had a harder problem than the others: it needed people who had never met to produce interchangeable assets. That constraint forced a level of specificity the rest do not reach.

The camera is given as a number. "The camera angle is top-down, roughly 60 degrees," with rendering "orthographic, which means there is no perspective." Two sentences that eliminate an entire category of unusable submissions.

The grid is fixed at 32×32 with optional 16×16 sub-tiles, and characters get their own envelope: the base fits 32×48, clothing fits 48×64. Anyone can check a submission against that without asking a question.

Its outline rule carries a deliberate exception worth noting, because most documents would have flattened it into one line and been wrong: outlines in general should be "a darker version of the current color, or a dark color generally, not black" — but for characters, "the outlines should be black or near black, no selective outlining." A different rule for the thing the player is actually looking at.

It closes with a section titled "Things that are important enough to be worth repeating": no pure colors, vary the hues of color ramps, highlights yellowish and shadows purplish, block out before detailing. A recurring-mistakes list, placed last, where people will actually reread it.

Battle for Wesnoth — Creating Unit Art

The shortest of the four and the densest in figures.

Wesnoth's document reads like a spec sheet because that is what it is. The canvas is 72×72 pixels, and one line fixes the baseline across every unit anyone contributes — which is the thing that makes a roster look like one roster rather than a pile of submissions.

The light direction argument, settled twice

The most quietly useful thing in the Wesnoth document is a parenthesis. Light comes from "the artist's (not the unit's!) upper right." LPC resolves the same ambiguity in the other direction — light primarily from above, any side directionality from the left.

Riot Games — League of Legends VFX style guide

The only one here from a large studio, and the only one about effects rather than static art.

Riot's guide is about a narrower problem — visual effects in a game where twenty of them can fire at once — but its rules generalize further than the topic suggests.

The part more documents should copy is the hierarchy rule. Within a single effect, one element is primary and defines the focal point, and secondary elements use color value and saturation to support it. When complementary colors appear together, "one color must act as the secondary color." That is a rule about the internal composition of one asset, and almost no indie art bible has anything equivalent.

Underneath all of it sits readability: effects must be "easily readable within gameplay" so players have "a clear sense of an effect's purpose." Same tie-breaker as 0 A.D., arrived at from the opposite end of the industry.

What all four have in common

They were written by different teams, in different decades, for a 3D RTS, a 2D contributor jam, a turn-based strategy game and a MOBA. They share five things.

Notice what is not on that list. None of them lead with mood boards. None of them open with adjectives about atmosphere. Three of the four get to a hard number within the first screen of text.

The other gap: interface art

The generated-art gap is the large one, and it has its own section above. There is a second, smaller one worth knowing before you copy any of these wholesale.

Only one of them has anything resembling an interface section. UI art collides with gameplay art constantly — a health bar that reads perfectly against a forest and disappears against snow is a UI problem discovered during a gameplay review — and its absence causes more rework than any other missing section. Riot's guide is about effects rather than interface, and the other three barely touch it.

Fill it in yourself rather than assuming the omission was deliberate.

How to use them

Do not read all four. Read the one closest to your problem, take its structure, and then steal the five common elements from wherever they are stated best.

Then measure your own art rather than adopting their numbers. Wesnoth's 72×72 canvas is correct for Wesnoth. The method — pick a canvas, state a baseline, write both down before the second asset — is what transfers.

And then add the section none of them has. Whatever route your art actually takes, write down what must not change between one asset and the next, and keep that list short enough to be checked in a minute. If a person is drawing, it is a list they read. If a generator is producing, it is a list something has to enforce on every render, because the generator will not remember on its own.

If you want to see what that looks like filled in for a specific game rather than described in the abstract, our art bible template is completed for a real project with measured values, and there is a blank version at the bottom of it. The Godot 2D art production spec is the delivery half split out on its own.

Frequently asked questions

Are there any public game art bibles?

Four are worth reading in full: 0 A.D.'s Art Design Document from Wildfire Games, the Liberated Pixel Cup style guide from OpenGameArt, Creating Unit Art from Battle for Wesnoth, and Riot's League of Legends VFX style guide. Most studios never publish theirs, which is why the open-source ones carry so much weight.

Do any of them cover AI-generated art?

None of them, and it is the largest gap in all four. Every rule in every one of these documents assumes a human makes each asset and can be given feedback, because the failure they were built to prevent is two artists disagreeing. A generated pipeline fails the opposite way: one source producing many internally-consistent outputs that quietly drift apart across a roster. That needs a section on what is held constant between generations — reserved colors, silhouette envelope, canvas and baseline, light direction, outline treatment, frame counts — and none of the four had any reason to write it.

Which public art bible is the most complete?

0 A.D.'s, at 21 sections covering theme, concept art, models, textures, animation, folder structure and naming. It is also the one whose structure translates most easily to a different project.

What do all good art bibles include?

A stated tie-breaker for when two rules conflict, a light direction with the artist-versus-character ambiguity resolved, shadow specified as a color value rather than an adjective, a level-of-detail floor tied to a real-world object, and a named list of the mistakes the team keeps repeating. If your art is generated rather than drawn, add a sixth: the list of properties that must not change between one generation and the next.

Should I copy one of these for my game?

Copy the structure, not the numbers. A 72×72 canvas and a 60-degree camera are correct for the projects that chose them. Measure your own art and write down what you find.

Why do these documents specify light direction so carefully?

Because "light comes from the upper left" is ambiguous the moment a character can face either way. Wesnoth resolves it as the artist's upper right, LPC as primarily from above with any side bias from the left. They disagree, and it does not matter — what matters is that each one settled it once in writing.