A giant planet that the James Webb Space Telescope could not clearly separate from its star's glare has now been pulled out of two-year-old archival images, not with new hardware but with new software. Researchers at Northwestern University reported in Nature on Nov. 4 that their method recovered 51 Eridani b, a young gas giant about 97 light-years from Earth.

The technique improved Webb's sensitivity to faint planets by 2 to 15 times compared with conventional image processing, according to a Northwestern news release. The gain varied by target, but the implication is broad: Webb's public archive may already contain planets that standard processing could not reveal.

That matters well beyond one star system. Webb is a NASA-led mission, and every hour of its observing time is a scarce, expensive resource. A method that extracts more science from data already collected stretches that investment without another launch.

The Starlight Problem Behind Hidden Planets

Directly photographing a planet is extraordinarily hard. "Planets are a million times fainter than their stars," lead author Rodrigo Ferrer-Chavez, a Ph.D. student at Northwestern, told Live Science.

Webb uses coronagraphs — masks that block most of a star's light — but they cannot block all of it. Astronomers typically subtract the remaining glare using a reference image of a different star. Tiny telescope movements and vibrations make that reference imperfect, leaving behind residual starlight that can bury even a large planet.

That is what happened with 51 Eridani b. Ground-based telescopes had already detected the planet, but in Webb's earlier observations, conventional processing could not separate it from leftover glare.

Building a Digital Twin of Webb

The Northwestern team took a different approach called FLIP, short for Physics-based Learning for Imaging Planets. Instead of borrowing another star's image, FLIP builds a digital twin of the telescope that simulates how starlight travels through Webb's optics to its detector.

The researchers adjusted millions of parameters to model small misalignments and optical distortions, then subtracted the simulated glare. The optimization tools resemble those used to train artificial intelligence, but the team stressed that FLIP is not an AI model and did not use AI to identify planets. "Because we already know the physics, we don't need AI to understand what's happening," co-author Jason Wang, an assistant professor of physics and astronomy at Northwestern, told Live Science.

The team tested FLIP on archival Webb data from its NIRCam instrument for three stars: HIP 65426, 51 Eridani, and TYC 5899, a star with no known planets. With FLIP, 51 Eridani b appeared at a detection strength above 5 sigma, the threshold astronomers commonly use for a confident signal.

The planet is a valuable target. Its atmosphere contains carbon dioxide, which absorbs light at 4.3 microns, a wavelength Webb can observe. That makes 51 Eridani b a good candidate for studying the chemistry of a young giant planet.

Old Data, New Discoveries

The most important takeaway may be what FLIP could find next. Most exoplanets known today are large, bright, or orbit close to their stars, partly because those are the easiest to detect. Ferrer-Chavez told Live Science that planets resembling those in our own solar system likely exist but remain hidden because current tools cannot see them.

Because FLIP works on data already collected, astronomers can rerun the method on years of Webb coronagraph observations. The team says the approach could help find smaller, colder planets without hardware changes.

It could also shape future missions. According to the Northwestern release, the technique may ease the extreme picometer-level stability that NASA's proposed Habitable Worlds Observatory would otherwise need. That observatory is intended to image Earth-like planets around Sun-like stars, and relaxing its stability requirements could affect its engineering and cost.

The work was funded by the Space Telescope Science Institute, the Alfred P. Sloan Foundation, and the U.S. National Science Foundation, according to the release.

Limits of a Single Recovery

The result should be read carefully. FLIP recovered a planet that was already known, not a new world, and the study tested the method on only three stars. The 15-fold improvement is the upper end of the range; the typical gain on other targets was smaller.

The method also depends on how well the digital twin matches the real telescope. Independent teams will need to apply FLIP to more Webb data, and to other instruments, before astronomers know how widely the gains hold. Any new planet candidate found in archival images would still require follow-up observations to confirm it is a planet and not background noise or a distant object.

Still, the study shows that better modeling of a telescope's quirks can unlock faint signals that were previously thrown away with the glare.

For readers, the significance is practical and long-term. Webb's archive is publicly available through the Space Telescope Science Institute, and methods like FLIP mean that past observations keep their scientific value. The next steps include applying FLIP to more stars, publishing new candidates, and testing whether the approach can reach the smaller, cooler planets that resemble Jupiter, Saturn, or eventually Earth.

The bottom line is that a software advance, not a new telescope, revealed a known giant planet in old Webb images and improved sensitivity by up to 15 times in some cases. Confirmation on more targets is still needed, but the method suggests that some of the next exoplanet discoveries may already be sitting in the archive.

What Readers Want to Know

What did the Northwestern team find? They recovered 51 Eridani b, a known giant planet, in two-year-old Webb images where conventional processing could not separate it from starlight glare.

What is FLIP? FLIP stands for Physics-based Learning for Imaging Planets. It builds a digital twin of Webb that simulates how starlight moves through the telescope, then subtracts that simulated glare.

Does FLIP use artificial intelligence? No. It uses optimization tools similar to those used in AI training, but the researchers say it relies on known physics and does not use AI to identify planets.

How much better is the new method? The team reported sensitivity gains of 2 to 15 times over conventional processing, depending on the target.

Could more planets be hiding in Webb's archive? Possibly. The researchers say FLIP could reveal smaller and colder planets in existing data, but any new candidates would need follow-up confirmation.

Where is 51 Eridani b? It orbits the star 51 Eridani, about 97 light-years from Earth, and its atmosphere contains carbon dioxide.