Black Hole Wallpaper: How We Simulated Gravitational Lensing in the Browser

Real gravitational lensing — the actual physics of light bending around a black hole — is a general relativity problem involving geodesics in curved spacetime. Solving that properly, per pixel, per frame, in real time on consumer hardware isn't happening. So this is the story of the shortcut, and why it holds up.

The shortcut

Instead of solving light paths through curved spacetime, the shader distorts screen-space UV coordinates radially around the event horizon, with the distortion falling off by roughly the inverse square of distance from center. It's a fragment shader doing a coordinate transform, not a physics engine doing ray tracing. Anywhere the background — the starfield layer behind the disk — gets sampled, it's sampled through that warped coordinate instead of the straight one.

That's the entire trick. It's cheap: one extra texture sample per pixel with a distorted UV, versus the actual physics which would need path integration per ray. On an M-series GPU this is well under a millisecond a frame.

The lensing warps the starfield behind the disk in real time. See Singularity in the marketplace →

Where it's honest, and where it isn't

What it gets right: the visual signature people recognize — the Einstein ring, the way stars behind the disk smear into arcs as they approach the horizon, the accretion disk appearing to wrap both above and below the hole simultaneously. That's real lensing behavior, and the radial-falloff shortcut reproduces it convincingly because human perception isn't checking the math, it's checking "does this look like the thing I've seen in photos of M87."

What it fakes: actual photon paths, gravitational redshift of light frequency, and the asymmetry a spinning black hole (frame dragging) would produce. The lensing is symmetric around the center; a real rotating black hole's isn't. I left this out — modeling rotation correctly would roughly double the shader complexity for a difference almost nobody would consciously register on a desktop wallpaper.

The accretion disk

Separately animated from the lensing distortion — a layered noise texture scrolling at different rates per layer, colour-graded from white-hot near the center to deep orange/red further out, which is a rough but recognizable stand-in for actual disk temperature gradients (real accretion disks are white-hot from Doppler beaming on the approaching side, which I also don't simulate — that's a genuinely asymmetric effect and a step further than felt worth it for a wallpaper).

Why cut it here and not elsewhere

Compare this to Ignis: Solar Wind, where the surface simulation is closer to physically accurate because the compute budget allowed it. Black Hole's constraint was different — the visual payoff of "correct" lensing math over "convincing" lensing math is small, and the CPU/GPU cost difference is large. Different scene, different place to spend the budget.

If you want the deeper architectural reasoning behind why this runs on a GPU shader instead of, say, a video loop at all: why we built Novaframe on OffscreenCanvas instead of video loops.

See it running in the catalogue.