Quantum Blur, Tessa Image and QPIXL run on the Moth Platform, and every job bills to this key. It is kept in this browser and sent with each request — never saved on the server. Get one at platform.mothquantum.com.
QPAM — amplitude-encodes each 256-pixel block into an 8-qubit state, applies an RX(theta) rotation that lets the amplitudes interfere, then reads the probabilities back as pixels. Higher theta = a softer, dreamier blur; Grain measures with fewer shots.
QSM — quantises every pixel into a 4-qubit value register alongside an 8-qubit index, and rotates all 12, spreading each value into a superposition that collapses into quantum static as theta climbs — a glitchy, posterised scramble.
Quantum Blur — maps the whole image onto a 2D Gray-code qubit grid and diffuses it with rotations, so neighbouring pixels bleed into one another. On the Moth Platform it is the blur-v1 engine; on Aer it is this app's own circuit, one per colour channel (19 qubits at 512px, 22 — the limit — at 2048px).
SQPAM / FRQI — entangles a position register with a single intensity qubit, then nudges every qubit together so their values interfere coherently. Yields a soft, ghostly light-leak shimmer that opens into abstract vertical light-bands at high theta.
NEQR — writes each pixel's value into 8 qubits with X gates and tips the chosen low bits into superposition, then measures once per pixel, so values randomly flip. Bit-depth picks how many planes go quantum: low = a fine film-grain wobble, high = big glitchy value swings.
FRQA — first applies a reversible operation to the whole image — invert (NOT gates on the value register), reverse (180° flip) or delay (a wrapping shift) — then NEQR's grain on top, with theta setting its strength.
QFT — runs the Quantum Fourier Transform on each block, entangles the resulting quantum-frequencies with a controlled-phase ring, then transforms back. Theta drives the band density (clean engraving → dense barcode) and Phi the entangling ghost/wash.
Quanvolution — slides a fixed, randomly-chosen entangling circuit (three layers of random RY/RZ and a CX ring) across the image, one 4-qubit circuit per pixel, reading one qubit back to carve an embossed bas-relief. Scale sets the contrast; Seed picks a different circuit. Up to 1024px.
Tessa Image — the platform's tessa-image-v1: every pixel becomes a point on a colour sphere (a radius and two angles — the same shape as a qubit), written onto a quantum device and read back. On the noiseless aer machine it comes back almost unchanged; the look comes from the noise of a real chip, or a model of one. At most 64×64 pixels, and far fewer on a chip.
QPIXL — the platform's qpixl-v1: pixel values encoded as qubit rotation angles (Interwoven QPIXL) and decoded from one measurement. Same size limits as Tessa; colour is one job per channel.
For the seven Aer effects, each block's pixels are loaded straight into the circuit as its starting state (Aer's set_statevector) rather than built from gates, and measurement shots are drawn from the distribution Aer returns with a seeded random generator — the same statistics as Aer's own sampler, which was too slow at image scale.