Posted by Every few months a batch of new space pictures goes around the world in a single afternoon — a nebula in impossible teal and gold, a star field so crowded it looks like spilled glitter. Almost nobody who reposts them knows they began life as a stack of grey rectangles. There is no colour camera on the James Webb Space Telescope. There never was one on Hubble either.
What comes down from an orbiting observatory is a grid of numbers. Each pixel counts photons for the length of the exposure and reports one value: brightness. Nothing about hue. Colour is added afterwards, on the ground, by people deciding which filter becomes which channel. That sounds like cheating. It isn’t, and the reason is more interesting than the accusation.
The Grey Rectangles Nobody Posts
A raw frame from Hubble’s Wide Field Camera 3 or from Webb’s NIRCam arrays is monochrome and, frankly, ugly. Cosmic rays punch bright streaks through it. Some pixels are dead, some run hot and report light that was never there. The detector’s own electronics contribute an offset before a single photon arrives. So every serious observation comes with a supporting cast of calibration frames: bias frames taken with the shutter shut and zero exposure, dark frames to measure what the chip invents on its own, flat fields shot against a uniformly lit surface to map which pixels are more sensitive than their neighbours.
Then the same target is shot again. And again.
Stacking is not just about prettiness. Signal accumulates in proportion to the number of frames while random noise grows only as the square root, so a hundred exposures buy you a tenfold improvement in the ratio between them. Telescopes also dither — nudge the pointing slightly between exposures — so that a bad pixel or the seam between detector chips does not land on the same piece of sky twice. Software later shifts the frames back into alignment, throws out the cosmic-ray hits that appear in one frame and not the others, and adds what remains.
Only now does anything resembling a picture exist. And it is still grey.
Why the Colours Are Representative, Not Fake
Filters are where the science lives. Broadband filters pass a wide slice of the spectrum; narrowband filters pass a slit of it, tuned to a single emission line from a single element. Ionised hydrogen glows at 656.3 nanometres. Doubly ionised oxygen radiates at 500.7. Singly ionised sulphur sits at 671.6 and 673.1. Point a narrowband filter at a nebula and the resulting frame is a map of where that one element is and how excited it is — a chemical survey disguised as a photograph.
The famous look of the 1995 Pillars of Creation comes from combining three such frames in what processors now call the Hubble palette: sulphur to the red channel, hydrogen to green, oxygen to blue. Hydrogen dominates most nebulae by a wide margin, so a naive natural-colour rendering would come out a muddy uniform pink and the structure would vanish. Splitting the lines across channels is what makes the ridges, cavities and ionisation fronts visible at all.
NASA and ESA stopped calling this “false colour” some years ago, and they were right to. The colours are assigned, but they are not arbitrary and they are not decorative — each one stands for a measurable physical quantity, and the recipe is published alongside the release. Order matters too: short wavelengths get blue, long ones red, preserving the intuition that blue means hotter or more energetic. A well-made composite is closer to a geological map than to a filtered selfie.
What Webb Sees That Hubble Cannot
Hubble, launched in April 1990 and rescued from its flawed mirror by a servicing crew in December 1993, works in the ultraviolet, the visible, and a little way into the near infrared. Webb starts roughly where Hubble gives up. NIRCam covers about 0.6 to 5 microns, MIRI runs out to 28, and to do that the whole observatory has to be cold — a tennis-court-sized sunshield holds the optics below about 50 kelvin, while MIRI’s own detectors are chilled to around 7.
Two things follow from working in the infrared. Dust that is opaque to visible light becomes translucent, so the interiors of star-forming clouds open up. And the expansion of the universe stretches the light of very distant galaxies out of the visible band entirely — ultraviolet emitted by the first generations of stars arrives here as infrared, which is why the earliest galaxies are essentially invisible to Hubble and routine for Webb.
Put the two versions of the Eagle Nebula pillars side by side and the difference is obvious. Hubble’s 1995 frame shows dense black silhouettes against glowing gas. Webb’s 2022 NIRCam version shows the same columns half-transparent, with newborn stars burning inside them and red lobes where jets from those protostars are punching into the surrounding cloud. Same object, same distance of roughly 6,500 light years, different physics on display.
Even the artefacts differ. Hubble’s four-pointed diffraction spikes come from the four vanes holding its secondary mirror. Webb’s eight come from the hexagonal edges of its eighteen segments plus its support struts. If you can read spikes, you can identify the telescope without being told.
Ten Days Pointed at Nothing
In December 1995 Robert Williams, then director of the Space Telescope Science Institute, spent a large block of his discretionary observing time staring at a patch of Ursa Major chosen precisely because it appeared empty. Colleagues thought it was a waste. Hubble took 342 exposures over ten days. The resulting Hubble Deep Field, in a field of view a few arcminutes across, held around three thousand galaxies nobody had known were there.
The Ultra Deep Field of 2003 and 2004 went deeper still — roughly a million seconds of exposure on a patch of Fornax about a tenth the apparent width of the full Moon, yielding some ten thousand galaxies. Webb’s first release in July 2022, the cluster SMACS 0723, reached comparable depth in about twelve and a half hours. Since then the JADES survey has pushed confirmed galaxies back to within a few hundred million years of the Big Bang, and the honest position among cosmologists is that some of these early systems look more massive and more mature than the models expected. Whether that means the models need adjusting or the mass estimates do is still being argued.
The Black Hole Picture Is Not a Photograph
The Event Horizon Telescope image of M87*, released in April 2019, is the odd one out. No camera was involved in any ordinary sense. Eight radio observatories from Hawaii to the South Pole observed simultaneously at 1.3 millimetres in April 2017, recording petabytes of raw data onto hard drives that were then physically flown to processing centres in Massachusetts and Bonn. The South Pole drives could not leave until the austral winter ended, months later.
Correlating those streams turns the Earth into a single telescope the size of the planet, sharp enough in principle to resolve an orange on the surface of the Moon. But the aperture is mostly holes — eight dishes, not a continuous mirror — so the sky brightness has to be reconstructed from badly incomplete data, and many different pictures fit the same measurements. The collaboration handled this by splitting into four teams that worked separately, with different algorithms, without comparing notes, and only then put the results side by side. They agreed: a bright asymmetric ring about 42 microarcseconds across, darker in the middle.
The dark patch is not the event horizon. It is the shadow cast by it, noticeably larger, on glowing plasma being whipped around a mass of roughly six and a half billion Suns. Sagittarius A*, our own galaxy’s black hole, followed in May 2022 and was harder — it is far smaller, gas orbits it in minutes, and the source changes while you are observing it.
The Data Is Sitting There, Free
Here is the part most readers of new space pictures never hear. Almost all of this material is public. Hubble and Webb observations flow into the Mikulski Archive at STScI, where proprietary periods are typically a year and often waived outright; ESA runs its own mirror. Planetary missions deposit into NASA’s Planetary Data System, and raw frames from Curiosity and Perseverance appear on mission pages within hours of downlink, unprocessed and unlabelled.
The files arrive as FITS, a format that has barely changed since the late 1970s and carries the full observing log in its header. Free software — Siril, AstroPy, GIMP with a FITS plugin — will open them. The community that has grown up around this is real and productive: Judy Schmidt has spent years mining the Hubble archive for objects nobody bothered to render, and STScI’s own Hidden Treasures contest in 2012 pulled genuinely striking results out of forgotten programme data. Some of the best-known versions of famous new space pictures were assembled not by mission staff but by people at kitchen tables.
Which is the whole point. The distance between the numbers coming off the detector and the image on your screen is a set of documented, repeatable choices, and anyone patient enough can walk it themselves.
If you would rather argue about palettes and pillars than scroll past them, SETIworld is where that conversation happens — new space pictures as they land, the processing behind them, and what they mean for the search for life elsewhere. Come and pick a target.
