i collapsed it into 1 pass over a Uint32Array, downsampled to ~45x45, reused canvases at module level, and added caching so changing baseSize/scaleFactor doesn’t re-scan
content detection: 1.4ms -> 37us per logo (38x)
full mount: 29ms -> 828us for 20 logos (35x)
github.com/sanity-labs/...
![for (let i = 0; i < pixelCount; i++) {
const pixel = data32[i]!;
// unpack RGBA from Uint32Array (little-endian)
const a = pixel >>> 24;
if (a <= contrastThreshold) continue;
const r = pixel & 0xff;
const g = (pixel >>> 8) & 0xff;
const b = (pixel >>> 16) & 0xff;
// squared euclidean distance from white
const distSq = (r - 255) ** 2 + (g - 255) ** 2 + (b - 255) ** 2;
if (distSq < contrastDistanceSq) continue;
// recover x,y from flat index
const x = i % sw;
const y = (i - x) / sw;
// accumulate weighted visual center
totalWeight += distSq * a;
weightedX += (x + 0.5) * distSq * a;
weightedY += (y + 0.5) * distSq * a;](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:7ajfpfi33pcwwq3i7xc7ova6/bafkreiaejfghmeqoxttbsmy3nr6i2ahq6svbgysmkug4342ppnynwawnee)
![Code block with the following code:
```groq
export const headingsProjection = \`
_key,
"title": array::join(children[].text, ""),
"level": select(
style == "h2" => 2,
style == "h3" => 3
)
\`
```](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:7ajfpfi33pcwwq3i7xc7ova6/bafkreianoymfv4444bsz3qaqkp4yssnj6typ5ldydl64kkzpimf545ldvy)
