A lens you hold against the world. Aim it at something and it finds the object under the aperture, takes that object’s own outline as the boundary of a vibrating membrane, and draws the standing wave it would carry — with the structure underneath layered into the same silhouette.
Chladni figures and electron orbitals are the same mathematics: standing waves of a wave equation under boundary conditions, with nodes where the amplitude stays at zero. So the figure is not decoration over the physics. It is the physics, at a scale you can point at.
It needs your camera to read the scene. Nothing leaves this device — the video never goes anywhere, and the whole instrument runs offline in this page. Without a camera it falls back to a simulated specimen.
drag to aim · pinch or scroll to sweep the frequency
Drag anywhere to move the lens. Pinch or scroll to sweep the dial. Clear takes the furniture away when you just want to look.
The top one is the dial: the ladder from the everyday world down into matter, nine rungs named along the top. Drag it slowly — each rung fades into the next, and the banner says which one you are on, how wide the view is and which rung of the nine. The second is the aperture: how much of the scene the lens covers.
The third is temperature, and it is the one control that sets something rather than reading it. Everything else here comes off the camera; a camera cannot tell how hot a thing is, so this is a what-if. Leave it on auto and it follows the temperature the instrument infers from how much the scene is moving. Drag it and that number is yours instead — the readout then says set rather than from agitation, and the things that really depend on temperature follow it: molecular speeds (√T), the air's density and the gaps between its collisions (both ∝ T at constant pressure), the lattice's thermal jitter, the Brownian walk of dust, and the thermal energy kT on the readout.
A chip is lit when it is on. Turning the others off is the quickest way to see what a single layer is doing.
Each rung draws its own layers, so a chip does nothing until the dial reaches a rung that uses it. Figure, Harmonics and Labels draw at every rung; Air and Kelvin are modes rather than rungs.
| Ambient | Field lines |
| Radiant | Field lines · Photons · Structure |
| Cellular | Field lines · Photons · Structure |
| Wavefront | Field lines · Wavefronts · Photons |
| Molecular | Field lines · Molecules (or Air) · Wavefronts |
| Atomic | Orbitals · Molecules |
| Nuclear | Orbitals · Nucleus |
| Partonic | Nucleus · Quarks |
| Vacuum | Quarks · Vacuum |
Until the lens knows what it is looking at, every object gets the same descent: generic cells, then a generic lattice. Name it with Identify and the two middle rungs become that material's own.
The wavefront rung changes with it, from light in general to what light does at that surface: wavefronts bending into glass and squeezed to λ/n inside it, with the 4% that turns back and makes a window faintly a mirror; turned away at the surface of steel, standing against their own reflection in nodes half a wavelength apart; walking their way out of paper a micrometre at a time, which is what white is; losing the blue and the red to chlorophyll in a leaf and coming back green; and leaving a screen in three narrow bands with nothing in between, which is what a screen's white is instead of a spectrum.
Every one of those numbers is a reference value for that kind of material, the same standing as the rest of the identity card — not a measurement of the thing in front of you. One thing does improve: where the material is known, the atoms drawn are the ones it is made of instead of the ones the pixel's colour would pick, and the caption says which of the two is in force.
It is the most asked question and the answer is a fact about the sensor, not a missing feature. Silicon stops responding to light at about 1.1 µm; the thermal radiation a warm body gives off is 8–14 µm, seven times further out, so a phone camera cannot see heat even in principle. Near infrared is within silicon’s reach but every phone bonds an IR-cut filter over the sensor to keep ordinary photographs from going pink, and ultraviolet is absorbed by the lens stack and the colour filters before it arrives. The Evidence chip shows the whole of what is left.
The Honest chip removes everything on this page that is a stated convention and leaves what was measured. The structure goes: the lattice and the material’s own micro-structure, which are reference values for a kind of thing rather than readings of this thing; the orbitals and the nucleus, which rest on colour → element; the quarks and the vacuum, which rest on pixel RGB → colour charge; and the air, which is drawn at an assumed density on a clock a million times the light clock. Four readings go with them: the photon energy and the electron de Broglie length, because both treat the light as a single spectral line and an RGB triple cannot say that; the scene clock, which is a stated slowdown; and the pitch, whose interval is exact but whose absolute value inherits the same assumption. A temperature you set goes; one read from agitation stays, which is why that row always says which of the two it is.
What is left is the light and the shape: dominant wavelength, photon flux, field amplitude, energy density, agitation, the field lines and wavefronts that follow measured light, the object’s outline, and the limits that follow from the magnification. The membrane stays and is labelled, because it is half of each — the silhouette is the object’s own, and the drumhead stretched across it is a model of something a real object is not.
Below the molecular rung the lens goes empty, and that is the point of the mode rather than a fault in it: everything drawn down there was drawn rather than read. The glass says what it removed and what the removed thing rested on, because an empty aperture on its own already means “nothing found”, and this is the opposite claim.
A camera measures light, not matter. Everything derived from light here is a real calculation on real data; everything about composition is a stated convention, applied consistently so that the same object always reads the same way.
Photon energy hc/λ, the electron de Broglie length h/√(2mE) and the pitch the Sound chip plays are all computed from the dominant wavelength as if the light were a single spectral line. It usually is not. An RGB triple cannot determine a spectrum — infinitely many spectra give the same colour, which is what metamerism means.
The clean case: sodium street lighting at 589 nm, and a red and a green LED mixed to the same yellow, read identically here. The instrument reports the same photon energy for both, and in the LED light there are no 589 nm photons at all. That is not an approximation, it is wrong, and it is listed here rather than quietly left in the readout. A diffraction grating over the lens would turn the guess into a measurement; the instrument does not have one.
The figure inside the lens is a damped wave equation utt + γut = c²∇²u solved on a grid clipped to that silhouette and driven at a frequency the dial sets. Bright filaments are the nodes — where sand collects on a real Chladni plate, because that is where the plate is not moving. The object’s shape is the boundary condition, so two different things under the same aperture at the same frequency give different figures, for the same reason two differently cut plates do.
Damping comes from how much the object is moving and how textured it is, so a still smooth surface rings cleanly and a busy one smears. The fold count of the harmonic line-work is measured from the outline, never chosen; a smooth blob scores almost nothing and the figure stays plain, which is the honest result.
The colour temperature is read from the brightest unclipped part of the frame and inverted with McCamy’s formula. Where the camera allows, its white balance is locked and then undone; otherwise the camera has already neutralised much of the colour and the reading is approximate. The spectrum is Planck’s law at that temperature — what the light would emit if it were an ideal black body. Real LEDs and fluorescents are spiky rather than smooth, so their UV and infrared shares are the black body’s, not theirs.
The Kelvin chip turns the lens into a map of that temperature, patch by patch, on a stretched scale so that 4000 K and 6500 K look different. Each patch’s colour is read as its light’s colour, which holds for white and grey surfaces; strongly coloured and very dark patches fade out instead of being guessed at.
Every rung centres on the crosshair: one atom of the lattice sits on it, that atom’s orbital is centred there, the nucleus sits at the orbital’s centre, and one nucleon at the nucleus’s centre holds the three quarks. Turning the dial goes into the thing under the crosshair: the rung you are leaving closes in on the centre while the next grows out of it.
Every rung is the same idea at a different size — a standing wave under a boundary — so the instrument lets them rhyme. The object’s fold symmetry and the direction it faces are measured from its outline; structure at every rung is turned to that direction and echoed with that fold count, in proportion to how strong the symmetry measured. Light-derived layers (field lines, wavefronts, photons) are never turned, because they follow measured light. The echo itself is a rendering choice, not a claim that atoms line up with the object.
Identify runs a MobileNet image classifier on whatever is inside the aperture. The model is served from this site and runs on the device; no picture is uploaded. It returns a guess over a thousand everyday categories, with its confidence, and it is often wrong — people in particular are not one of its categories, so a person comes back as clothing.
The material facts it then shows are looked up from a short written table by keyword: what that kind of thing is typically made of and how big its structure is. They are reference values, not measurements of the object under the lens, and nothing in them comes from the network.
With nothing separable under the aperture, the subject is the volume of air itself: 2.5 × 1025 molecules per cubic metre, 3.4 nm apart, 68 nm between collisions, moving at about 470 m/s. At the molecular rung those are drawn at that density and at their real size, which is why so few of them cross the view: a gas is mostly nothing.
Two honest cheats, both stated on the captions: the view is a slab of the stated depth, because a full column of air holds far more molecules than can be drawn; and the molecules run on their own clock, a million times the instrument’s light clock, because against light they are effectively still.
Each molecule also collides: the distance to its next collision is drawn from the real distribution for a 68 nm mean free path, and its speed afterwards from the Maxwell–Boltzmann spread around 470 m/s. Most partners are outside the drawn slab, so a collision shows as a kink and a flash. On the gas clock that is one every 145 ps per molecule — no further speed-up is needed to see them.
At the radiant rung the air carries its dust: about 3 × 107 specks per cubic metre of ordinary room air, most of them small, looked at through 60 cm of sunbeam. Big specks fall at the speed where air drag balances their weight (Stokes’ law — about 4.5 mm/s for a 10 µm grain); small ones barely fall but jitter, knocked about by the air molecules (Brownian motion). This is the one thing in the lens shown at its true speed.
At the wavefront rung the air shows why the sky is blue. White light crosses the lens as photons of every colour, each scattered with a probability that goes as 1/λ4: blue at 450 nm scatters 5.9 times as often as red at 700 nm, so the scattered light is blue and the beam that carries on reddens. In real air visible light travels about 90 km before it scatters, so the rate is boosted by the factor the caption prints; the ratio between colours is the real one. The light enters from the brightest thing in view outside the lens.
Until something is identified the ladder is nominal: its top rung is a metre across because a number was needed. Identify something whose kind is in the table and the ladder is anchored instead — the object’s measured width in pixels against the typical size of that kind of thing gives metres per pixel, and every field of view, pixel size and drawn distance below it follows. The readout names what the anchor came from, and it remains an assumed size for that kind of object, not a measurement of this one. Correcting the label re-anchors it to whatever you chose.
build —
+ − sweep · 1–9 jump to a rung · [ ] aperture · space freeze · f flip camera · s capture · p readout · r recentre lens