Crisp Material Transition Design¶

Purpose¶

This note captures an experimental direction for improving zoom-detail material transitions in the terrain renderer.

The current terrain style is not photorealistic. At zoomed-out and mid-zoom levels, the top-down terrain map has a visible pixel-art character. At close zoom, material detail maps add high-frequency texture. Smooth linear blending between material details can feel muddy or "wishy-washy", especially when the base terrain is already stylized and pixel-sharp.

The goal is to explore material transition methods that preserve a sharper, more intentional visual style.

Problem With Smooth Detail Blending¶

A standard material splat blend mixes material colors continuously:

color = dirt * wDirt + grass * wGrass + rock * wRock;

This is simple and stable, but it can cause issues for this project:

Material identity becomes unclear in transition zones.
High-frequency detail maps smear into each other.
Transitions can resemble older muddy terrain blending.
The result can conflict with the pixel-art-like macro terrain style.

The base splat/albedo map can remain smooth or painterly. The concern is mostly with the close-zoom detail material layer.

Direction: Dithered Material Selection¶

Instead of blending all material detail colors continuously, use splat weights as probabilities and choose one material per fragment, map pixel, or detail sample.

For two materials:

float n = noise(mapCoord);
float useGrass = step(n, grassWeight);
vec3 material = mix(dirtColor, grassColor, useGrass);

If grass has a weight of 0.4, roughly 40% of samples become grass and 60% remain dirt. The transition is spatial and crisp rather than blurry.

For multiple materials, use cumulative normalized weights:

float n = noise(mapCoord);

if (n < w0) {
  material = mat0;
} else if (n < w0 + w1) {
  material = mat1;
} else if (n < w0 + w1 + w2) {
  material = mat2;
} else {
  material = mat3;
}

This produces a stippled or dithered material boundary while preserving hard material identity at each sample.

Noise Source¶

The dither/noise source should be stable in map space. Screen-space noise would shimmer while panning or zooming.

Candidate sources:

Blue-noise texture tiled in map space.
Small ordered Bayer matrix in map pixel space.
Hash noise from integer map/detail coordinates.

Blue noise is likely the best visual default:

Less clumpy than white noise.
Less visibly grid-like than Bayer dithering.
Good for stylized and pixel-art-adjacent transitions.

A small 64x64 or 128x128 blue-noise texture should be enough for a first experiment.

Example map-space sampling:

vec2 noiseUv = floor(mapPixelCoord) / 64.0;
float n = texture(uBlueNoise, noiseUv).r;

For close-zoom detail, the noise could also be sampled in detail texel space if finer transition granularity is desired.

Preserve Base Terrain Coherence¶

The base terrain color map should remain the macro source of coherence.

Recommended split:

Base terrain splat/albedo remains as authored.
Detail material is selected crisply with dithered material choice.
Selected detail is mixed into the base color using the existing zoom fade.

This keeps large-scale terrain color stable while making close-zoom material identity sharper.

Dominant Plus Accent Selection¶

Pure weighted random selection can become noisy if every small nonzero material weight contributes visible speckles.

A better mode may be dominant-plus-accent:

Find the dominant material.
Allow secondary materials mostly near transition zones.
Use noise to place secondary material patches.
Suppress extremely small weights unless they exceed a threshold.

This preserves readable regions while still avoiding muddy blends.

Example behavior:

80% dirt / 20% grass: mostly dirt with sparse grass patches.
52% dirt / 48% grass: strong transition pattern.
95% dirt / 5% grass: almost pure dirt.

Quantized Weights¶

Quantizing material weights can reinforce the stylized look.

Example:

w = round(w * 7.0) / 7.0;

or:

w = round(w * 15.0) / 15.0;

This creates stepped, intentional transitions and pairs well with dithered selection. It can also emulate lower-precision material encoding before building actual packed splat textures.

Priority Dither Without Material Height Maps¶

Height-based blending often looks good, but it requires per-material height maps. To avoid that extra data cost, approximate the effect with material priority values.

Example idea:

score = weight + materialPriority + noise * noiseStrength;

Then choose the material with the highest score.

This allows authored transition tendencies:

Rock breaks through grass.
Snow sits over rock.
Dirt fills low-priority gaps.

Example priorities:

rock +0.05
grass +0.00
dirt -0.02
snow +0.03

This is not physically correct, but it may produce more intentional-looking transitions than pure probability.

Candidate Detail Blend Modes¶

A useful experiment would expose three modes in the detail material system:

Smooth Current weighted material blending.
Dithered Select one material using blue-noise cumulative weights.
Priority Dither Select one material using weight, material priority, and noise.

The third mode may fit the project best because it keeps transitions sharp while allowing art-directed material behavior.

Parameters To Expose¶

Potential controls for dev tuning:

Detail blend mode: smooth, dithered, priorityDither.
Dither strength.
Transition noise scale.
Weight quantization steps.
Per-material priority.
Minimum visible material weight threshold.

These should initially live in the existing zoom-detail material panel rather than becoming a separate system.

Packed Splat Texture Note¶

Packing two material weights into one 8-bit channel is possible only with up to 16 levels per material:

16 * 16 = 256 combinations

A natural encoding is:

packed = a + b * 16

a = packed % 16
b = floor(packed / 16)

This can represent eight material weights in a single RGBA texture if each material uses 0..15 precision. It should be tested visually before committing, because 16 levels may cause banding in smooth transitions.

An alternative is to use two RGBA splat textures for eight materials at full 8-bit precision. This costs another texture lookup but preserves blend quality and simplifies authoring.

Recommended First Experiment¶

Implement the visual test before changing asset formats:

Add shader-side weight quantization for current four-material detail splat.
Add dithered material selection using map-space blue noise or hash noise.
Compare smooth, dithered, and priorityDither modes.
If dithered transitions look good, evaluate packed 4-bit material weights.
If banding/noise is too visible, prefer a second splat texture over packed precision loss.

This keeps the experiment cheap and reversible while directly testing the visual question.