Any shape

Rather than typing the planform in, let the package measure the model:

const { scene } = useGLTF("/jet.glb");

<group>
  <primitive object={scene} />
  <WingVapor
    capture={scene}
    // Not the shape the air flies round
    captureFilter={(mesh) => !/Gear|Pylon|Missile/.test(mesh.name)}
    flight={flight}
    air={air}
  />
</group>;

What it measuresCopy Anchor Link

capture takes six orthographic depth views of the model, from above and below, ahead and astern, and either side, at 128 pixels along its longest side. They're rasterised on the CPU, so they need no renderer and come out the same everywhere: tens of milliseconds for a few hundred thousand triangles, once. From them it measures:

• Every station of the wing across the span: its leading and trailing edges, its thickness and its mid-plane's height. A cranked delta keeps its crank. A tailplane or canard at the same stations is told apart from the wing by the clear air between them, and a fin standing on the wing by how tall it is.
• Where the wing meets the body: where the leading edge leaps forward to the nose, or the section gets far thicker for its chord.
• The spanwise loading, from a Weissinger vortex lattice on the measured planform.
• A canard or a tailplane: the second run along the chord, ahead of the wing or behind it. Its own tips shed their own vortices. A canard carries about 1.2 times its share of the area in lift, so it trails vapor in a hard pull. A tailplane only trims, at about 0.15 times its share, and rarely fogs.
• The body, as the round body of the same largest cross-section, for the cone.

The field reads the wing as that table, station by station, so the sheet follows the real planform and the tip vortices leave from the real tips. The measured planform is also fitted with straight edges into airframe, for the lift and breakdown correlations. airframe still wins for whatever you give it.

The aircraft's mass and how sharp its leading edge is can't be seen in depth views, so they stay as given.

What to leave outCopy Anchor Link

Hide the landing gear and the stores, or leave them out with captureFilter: they are not the shape the air flies round. Engine nacelles and pods jutting from an edge are cut off by themselves: anything narrower than about a seventh of the semispan that juts from an edge is not the wing the air lifts on.

Depth views only see the outside, which is what the air sees too: an intake duct changes nothing a cloud forms in.

In the page, without a stallCopy Anchor Link

<WingVapor capture> measures a few milliseconds a frame, so a big model doesn't drop frames. Until it is done the vapor flies airframe as given.

Without React:

const measured = await vapor.captureAsync(model, {
  filter,
  budgetMs: 4, // per slice
  signal: abort.signal,
});

capture_airframe_async does the measuring alone. vapor.capture and capture_airframe do it all at once.

Ahead of timeCopy Anchor Link

Or measure at build time, and ship the numbers in place of the capture:

// A build script, in Node: capture_airframe needs no renderer
const baked = serialize_capture(capture_airframe(model, { filter }));
writeFileSync("jet.vapor.json", JSON.stringify(baked));

// At run time
import baked from "./jet.vapor.json";

<WingVapor measured={deserialize_capture(baked)} />;

A bake is versioned: one from a package whose table has changed is refused, not misread.

On real modelsCopy Anchor Link

Besides the docs' fighter, the capture was checked on two models it was never tuned on: Cesium's twin turboprop and Babylon.js's aerobatic plane. The twin measured 25.7 m across, its wing 4.2 m deep at the root and 0.8 m at the tip, 7° of sweep, the body 2.25 m out, and a tailplane 5.6 m across. Its nacelles were what taught the capture to cut pods off. Both fly as you would expect: a sheet over the wing in a hard pull in humid air, and no tip trails. A 900 kg aerobatic plane at 7 g sheds a vortex a tenth as strong as the fighter's, too weak to fog in 95 % air.

See it on aircraft made of primitives.