The Milky Way over Maunakea, and six more cameras worth the dark
At 03:08:41 Hawaii time this morning the all-sky camera at the Subaru Telescope had the Milky Way lying corner to corner, the Subaru enclosure squat at the bottom of the fisheye and the two Keck domes pale on the left. Printed into the corner of every frame are the conditions it was taken in. What I fetched at 13:08:41 UTC on 2026-10-16 says sun elevation -44.959 degrees, moon elevation -67.605 degrees, moon 30.700% illuminated, which is the astronomer's way of saying the sky is as dark as Maunakea gets.
Most cameras lose their subject at sunset. A few of them get one.
Three ways to survive the dark
A camera that is worth opening after dark is doing one of three things. It is pointed at its own light, like a night market or a lit bridge. It is pointed at light that makes itself, which in practice means lava or stars, with the aurora as the special case. Or it was built with a night exposure, in which case it is usually bolted to an observatory roof and pointing straight up.
The third kind is the easiest to check, because the observatories publish their sky cameras and timestamp every frame. Gemini North, on the same summit ridge, runs an upward view whose current frame was stamped 2026-10-16T02:59:49-1000 when I pulled it this afternoon: a grainy, brown-black star field with four or five bright stars throwing diffraction spikes in the top left corner. It is a working instrument rather than a pretty one. You are looking at what the telescope operator looks at when deciding whether to open.
Izaña, AEMET's observatory on Tenerife, solves the exposure problem by not solving it. It publishes two all-sky images from the same roof, one exposed for daylight and one for the dark, and they hand over at dawn and dusk. The day camera was stamped 2026/10/16 13:10:01 when I fetched it, a blue-grey lid of cloud over the whole island. The night camera was stamped 06:00:01 that same morning and will stay there until this evening: cloud underlit orange by the towns below, a scatter of stars through a gap, the observatory's own buildings black around the rim of the lens. That last frame of the night then sits on the page all day like a held breath.
Our day and night pages for it are separate cameras, because upstream they are separate cameras.
Sakurajima, with stars
The Japan Meteorological Agency keeps cameras on the active volcanoes, and the one named for Ushine, on the Tarumizu shore, looks across the water at Sakurajima. At 22:01:55 Japan time on 2026-10-16 the mountain was a flat black cut-out against a sky with real stars in it, a thin plume leaning off the summit to the right, the lights of the coast road in a string at the bottom of the frame and a lamp standard in the left foreground. The exposure is long enough to hold the stars and still show the plume, which is a harder trick than it sounds.
The ones lit by the city they are in
Korea's traffic cameras are the other half of this. Gwangan Bridge, upper deck, marker 23 looks along the upper deck, lane markings painted for Seomyeon and Namcheon, with the Gwangalli waterfront lit behind it. The still our server was holding at 13:12 UTC had two cars on an empty deck and the tower blocks burning away across the water. Street lighting is constant and close, so these cameras read much the same at 2 a.m. as at 2 p.m.
Which is literally true of the tunnel cameras. Yunsan tunnel at Hwamyeong looks identical whenever you open it: green-white strip lights down a curved concrete bore, red and white lane markings, a green exit sign on the right, one car in the frame I opened. A tunnel camera has no idea what time it is, and that is its charm.
Taiwan's Luodong night market is the version of this I would actually open. The frame I pulled at 13:10 UTC carried a burnt-in 10/16/2026 20:47:06: stalls with strip lights under awnings on the left, a dozen scooters waiting at the line, a white hatchback halfway through the junction, people crossing in both directions. The second camera, about 140 metres north at Minsheng Road, was stamped 20:31:06 and had two dozen or so people in the crossing at once. These come from a snapshot host rather than a live stream, so what you get is minutes old: the files behind both cameras were last modified at 12:51 and 12:59 UTC when I checked the headers at 13:19.
What the dark does to everything else
Two Makaha Ridge cameras on Kauai came second and third in the most-viewed pages on the site yesterday, on 86 and 85 views, counted from our own page-view digest for 2026-10-15 with crawlers excluded. At 03:03 Hawaii time this morning the northwest one was a grey-black rectangle with perhaps twenty faint stars in it and a hint of a horizon two thirds of the way down. FAA weathercams are for pilots, and a pilot wants them at dawn. That is the honest shape of the night side: a few cameras gain, a lot of them wait.
The aurora cameras are the exception that earns its own post, and the hour-by-hour list of those is here.
What I want to build next is the obvious thing. We hold coordinates for every camera, and the frames carry their own clocks, so the sun's elevation where a camera stands is something we can work out rather than guess. Put that next to the refresh interval on the camera page and the map can be sorted by what is dark right now, which is how I have been picking cameras by hand all week.
The part I have not solved is ranking them. Lit and visible is not the same as worth watching, and I do not yet have a measure that separates the Luodong crossing at nine in the evening from an empty tunnel at the same hour. If you know a camera that is better at 2 a.m. than at 2 p.m., send it and I will go and look.