Feature Library

Coverage-keeping atlas mipmaps

0.15 GPU · 0.18 CPU ms
A still of the Coverage-keeping atlas mipmaps demo
Still
The live demo starts when it’s in view.Compare plain and corrected mipmaps · change distance or cutoff · Near view and Far view keep both cameras matched

Build explicit RGBA mipmaps for alpha-tested atlases so thin leaves, stems and fence wire survive minification.

Each cell targets its own base coverage; alpha-weighted RGB avoids transparent-colour fringes, and an uncorrected pyramid prevents opacity boosts from accumulating. Pure CPU preprocessing with no renderer dependency.

What it costs a frame

0.15 ms
0.18 ms
GPU
0.15 ms
CPU, median
0.18 ms
CPU, 95th pct
0.30 ms
Draw calls
4
Weight
Light

Measured 19 November 2026 on RTX 4060 Laptop, 13th Gen Intel(R) Core(TM) i9-13900H, Chrome 157, 1920×1080 at 1×. 600 frames × 3 rounds, median kept (rounds ranged GPU 0.06–0.18, CPU 0.17–0.20 ms). What ran: Two synchronized views of one code-generated 256×256 RGBA atlas on identical quads. Left uses an uncorrected box-filter chain, right uses per-cell coverage correction; both use alpha-weighted RGB, alpha test and trilinear filtering. Mips build on load/cutoff edits, not per frame. No automatic motion, external textures, blending, alpha-to-coverage or lighting. Phones: tiers are being measured next.

How steadily it runs on RTX 4060 Laptop · 13th Gen Intel(R) Core(TM) i9-13900H · reference

Typical frame ?
3.10 ms
1% low ?
149 fps
Varies by ?
±1.21 ms
All metricsRTX 4060 Laptop · 13th Gen Intel(R) Core(TM) i9-13900H · reference
p50p95p990.13 ms7.14 ms +
How long each frame took. Most frames sit in the tall bars; a long tail to the right is hitching.

Frames run back to back, each starting once the GPU has finished the last, so frame times show the work itself. They have a floor of about 4–5 ms from the browser’s own scheduling: GPU and CPU ms stay the measure of cost.

Frames

Median (p50) ?
3.10 ms
p95 ?
5.26 ms
p99 ?
6.14 ms
Slowest ?
7.32 ms
Mean ?
3.25 ms
Standard deviation ?
1.21 ms
1% low ?
149 fps
0.1% low ?
137 fps
Hitches ?
4 of 600
Over budget ?
0 of 600

Main thread

Long tasks ?
0
Long frames ?
0
Event-loop delay ?
0.0 / 0.3 / 0.3 ms

Input

No input during this run.

Memory

JavaScript heap ?
41.4 → 42.7 MB
Growth ?
+1.33 MB

Browser

JavaScript a frame ?
0.01 ms
All tasks a frame ?
0.37 ms
Layout / style ?
0.00 / 0.00 ms
Main thread busy ?
11%

Garbage collection

Collections ?
5 (5 minor, 0 major)
Pauses ?
1.6 ms in all, longest 0.3

What it needs

Renders with
WebGPU or WebGLWebGPU where there is one, WebGL2 where there is not.
Your browser
Runs hereNo WebGPU needed
Depends on
three.js 0.187.0No other packages
With an agent
install("alpha-mips")Your coding agent reads its pitfalls and installs it through the library’s MCP server. Set it up

What it’s for

  1. Keep thin alpha-tested foliage visible at a distance
  2. Preserve sparse fence wire through mip reduction
  3. Build per-cell atlas mipmaps without loading source art
  4. Avoid transparent RGB fringes and cumulative opacity inflation

Pitfalls, in brief

  1. Input is tightly packed unpremultiplied RGBA8 in a Uint8Array, with positive power-of-two width/height and square power-of-two cells fitting both dimensions.
  2. cutoff is a byte threshold from 1 to 255, not a normalized 0–1 value.
  3. Coverage is a heuristic, not exact area conservation: rounded texel counts, tied alpha values and the 1.5× cutoff headroom can overshoot.
  4. Cells remain independent only while each is at least one mip texel.
  5. Never feed a corrected mip back into the next average.
5 more in brief
  1. RGB is alpha-weighted in the supplied byte space, not automatically linearized.
  2. The returned array includes the original base object at level zero; its bytes are not modified.
  3. Upload the whole explicit chain including level zero, set generateMipmaps=false and a mipmapped minification filter.
  4. This is for alpha-tested cutouts, not ordinary alpha blending.
  5. Build on load or asset/cutoff edits, never each frame.
10 pitfalls in full, the hook-up guide, the module source and its tests come with the part.

Where it comes from

From HaTac. Adapted from published work, named below.

Hannes Andersson: Source per-cell coverage mip builder, uncorrected pyramid and atlas integration.

alpha testmipmapsatlascoveragefoliageleavesfencescutoutstexture filteringCPU

Near it in the frame