feat(registry): add the halftone-field texture block

[miguel-heygen]

What

halftone-field: a grid of dots whose size and colour track a flowing noise field, so broad diagonal ridges sweep across a mostly-black frame.

Why

The catalog has no texture register at all. This is a background that runs under type without competing with it, and it gives us a general "scalar field to halftone" pass to point at other content later.

How

Closed form by construction. The field is a pure function of position and time, so the entire port is uTime = frame / fps: no accumulator, no PRNG, no history anywhere.

One pass rather than the reference's two. Its first pass is analytic, so evaluating the noise at each cell centre returns exactly what a texture fetch would. The render target only earns its place if the field ever becomes something that cannot be evaluated closed-form (a real texture, a feedback buffer).

Repaint is driven from a property setter on the tweened driver rather than a timeline onUpdate callback, because GSAP suppresses events on seek by default and a canvas driven that way freezes on frame 0.

The two defaults that depart from the reference

Worth reading before assuming they are taste. The reference's own arithmetic contradicts its stated look.

At paletteBias = 10 the bias is 0.5. Since gray = luma^4.84 lands in [0, 0.70], the value driving both dot radius and palette position is pinned to [0.5, 1.0]. Two consequences:

• every dot has a hard floor of 0.25 cells, so nothing can ever approach black
• the palette window covers only stops 2 to 4, which makes #00AAFF arithmetically unreachable

Rendered, that is a flat evenly-lit purple-grey lattice: the same failure the gamma is supposed to prevent, arriving through bias instead.

Every formula and mapping is kept verbatim. Only two dial defaults move, paletteBias 10 -> 2 and frequency 5 -> 2, and both are one slider from the reference values.

Test plan

• Unit tests added/updated
• Manual testing performed
• Documentation updated (if applicable)

hyperframes check passes with 0 errors: layout 0 issues across 9 samples, motion 0, contrast 0. Registry item lint 0 errors. oxfmt clean. The runtime warnings are llvmpipe ReadPixels GPU-stall notices from software rendering.

WebGL surviving seek-capture was the main risk, so it was tested rather than assumed:

seek    -> t=3.0 (run 1): 3f8a92ccb2c4...
seek    -> t=3.0 (run 2): 3f8a92ccb2c4...
stepped -> t=3.0        : 3f8a92ccb2c4...

Direct seek equals stepping 90 frames equals a fresh run. Two CLI snapshot --at 4.2 runs are byte-identical. Frames were rendered and looked at at 0/2/4/6/8/9.9s: the main ridge travels lower-left to centre to upper-right, movement is already visible at 0.25s, and the troughs stay near-black with real gaps between dots.

The quantise toggle has its own check: stepped t=3.000 equals t=3.020 (same 30fps bucket) and differs from t=3.050, while smooth mode separates all three.

Not covered

Frame 0 is already lit rather than black. "Near-black to a blaze in under a second" describes the band sweep, not an intro, and a scripted fade would contradict the closed-form contract. An actual entrance is a small addition if wanted.

No catalog preview PNG; the mdx points at a URL that 404s until scripts/upload-docs-images.sh runs with the publishing profile. Maintainer step.

The file trips the 300-line advisory warning at 348 lines, about 90 of which are the simplex-noise function. Not splitting that.

[mintlify]

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated (UTC)
hyperframes	🟢 Ready	View Preview	Aug 8, 2026, 4:22 PM

💡 Tip: Enable Workflows to automatically generate PRs for you.

[miguel-heygen]

Superseded by #3116, which carries all five video primitives in one PR as requested. Same commits, same verification; nothing dropped.