A telescope can collect the light from a distant planet and still fail to show the planet in its final picture.
That sounds like a contradiction. If the light reached the detector, shouldn’t the object be visible? Not necessarily. In astronomy, what gets recorded and what scientists can reliably separate are two different things. A star’s light can spread across an image in complicated patterns, overwhelming a nearby planet that is thousands or millions of times fainter.
On November 4, 2026, researchers reported in Nature that a new analysis method had recovered the signal of a previously known exoplanet, 51 Eridani b, from existing observations by NASA’s James Webb Space Telescope (JWST). Earlier processing had not detected the planet in one of those infrared images.
The telescope had not changed. The scientists changed how they separated a planet’s light from a star’s glare.
Why a photograph of a star can hide a planet
Consider photographing a firefly beside a stadium floodlight. The floodlight does not merely appear as a bright point. Light leaks across the optics, reflects, scatters and produces patterns that can bury the firefly.
A planet next to its host star poses a similar problem, except that the brightness contrast can be enormous. Astronomers use an instrument called a coronagraph to block much of the star’s direct light. But the remaining light is not perfectly clean. Tiny imperfections in mirrors, detectors and alignment create a pattern known as the point-spread function, or PSF.
That residual pattern is more than an inconvenience. It can look like a collection of fake sources, and it can conceal a real one. Even a remarkable telescope such as Webb has to contend with the physics of its optics.
Astronomers have developed sophisticated ways to estimate and subtract the unwanted starlight. But if that estimate is even slightly inaccurate, subtraction can erase a planet, leave false spots behind, or make a faint object statistically unconvincing.
The new tool was named FLIP
The researchers call their method Physics-based Learning for Imaging Planets, or FLIP. Instead of treating the unwanted starlight simply as a pattern to remove, FLIP builds a physical model of the telescope and asks how that instrument would actually produce the pattern observed.
The method uses information about the telescope’s optical state, then adjusts model parameters using automated differentiation and numerical optimization. Those parameters include how precisely the star is positioned behind the coronagraph, small changes in the optical wavefront, and detector response.
In plain language, FLIP makes a best-possible computer version of the star’s glare, checks how it differs from the real observation, and refines the model until it can subtract more of the unwanted signal without destroying the faint companion nearby.
This is a clever combination of physics and computation—not simply a generic instruction to artificial intelligence to ‘make the image clearer’. The model is constrained by how a real telescope works.
What emerged from the archive?
The team tested FLIP on real observations of several star systems. Among them was 51 Eridani, a nearby star with a giant planet already discovered using ground-based instruments. That planet is called 51 Eridani b.
The crucial observation came in Webb’s near-infrared camera, NIRCam, using its F430M filter around 4.3 micrometres. Conventional image processing had not identified the planet in the original analysis at that wavelength. With FLIP, the researchers obtained a convincing detection in the same observations.
That distinction must not be lost: this was not the discovery of a new exoplanet. It was the recovery of a known planet in a previously unsuccessful wavelength measurement. That is scientifically valuable because different infrared wavelengths carry different clues about planetary atmospheres.
Near 4.3 micrometres, carbon dioxide strongly absorbs infrared light. Improving observations in this range can eventually help astronomers understand atmospheric composition. The new detection, however, is not by itself a complete carbon-dioxide abundance measurement; further analysis is needed.
How much better was the new analysis?
In the peer-reviewed paper, the researchers report that their framework improved detection sensitivity by factors of roughly 2 to 15, depending on the data and location in the image. That range is important: fifteenfold is the most impressive result in the tested conditions, not a promise that every Webb picture becomes fifteen times sharper.
At the location of 51 Eridani b in the F430M data, the team estimated an improvement in sensitivity of about four times compared with standard processing. They also tested the method with other known planetary signals and simulated planets added to actual images.
A strong method has to do more than make a picture look attractive. It needs to preserve the positions and shapes of genuine signals and avoid manufacturing detections from noise. The researchers compared their results with existing methods and performed tests in which artificial planets were inserted into data to see whether they could be recovered.
There are still caveats: an algorithm can produce artifacts, some image regions remain uncertain, and what works in one instrument or observation mode may need careful adaptation elsewhere.
Why does this matter for future telescopes?
Building a telescope capable of directly photographing a small planet beside a bright star requires extraordinary optical stability. For proposed missions aimed at Earth-like worlds, tiny variations in the instrument can become a formidable obstacle.
One response is to build a steadier, more precise telescope. Another is to improve how scientists understand and correct the imperfections that remain. The future will almost certainly require both.
FLIP suggests that better models of the instrument can help recover signals even when the hardware is not perfectly stable. Researchers see potential applications to future observing systems, including NASA’s planned Nancy Grace Roman Space Telescope and concepts for a Habitable Worlds Observatory.
That is a promising direction, not a guarantee that scientists can solve every hardware problem with software. The study demonstrates performance on particular Webb datasets. A broader role would require tests on other instruments and more diverse observations.
Could older telescope data hold discoveries we missed?
This is the question that makes the work larger than one planet. Modern observatories generate an enormous archive of observations. Once those photons have been recorded, improved methods may let researchers ask new questions of the same data years later.
Sometimes a new discovery requires a new spacecraft. Sometimes it requires looking again at an old image with a deeper understanding of the instrument that made it.
The 51 Eridani result belongs to the second category. Astronomers did not suddenly move the planet into view. They learned how to see more clearly through the glare that was already there.
And that raises an irresistible possibility: not every invisible world is waiting for a future telescope. Some may be waiting for a better way to read the images we already have.
Sources and research notes
- Ferrer-Chavez, R., Wang, J. J., Bouman, K. L., et al. (2026). Imaging exoplanets below the telescope stability limit. Nature, 659, 58–62. Published November 4, 2026. Primary peer-reviewed, open-access paper.
- Nature. Full paper and Figure 2 comparison. The scientific image panel compares conventional KLIP processing with the FLIP reconstruction; publication is CC BY 4.0 except where third-party credits indicate otherwise.
- Northwestern University (2026). Hidden planet emerges from the glare. Research news release, including the downloadable original image comparison.