The popular media article is describing this PRL paper
All-Optical Delay of Images using Slow Light
by Ryan M. Camacho, Curtis J. Broadbent, Irfan Ali-Khan, John C. Howell. Check the arXiv version that there is the "UR" image on the last page of the paper.
The achievement is that the optical pulses lasting 2 ns may be delayed by 10 ns using some clever interference between the delayed images and local oscillators.
The popular article's focus on the "information stored in a single photon" is utterly misleading. A single photon only carries 3 continuous degrees of freedom we may choose, e.g. the momentum vector, and 1 qubit of information, e.g. the linear $x/y$ or the circular L/R polarization. This is clearly insufficient for a natural encoding of an image.
The actual point of the paper related to the "single photon" was that the interference used to reconstruct the delayed images only relied on the wave character of "whatever represents the light". One may derive that it works for a classical electromagnetic field. But the same gadget inevitably works even if the electromagnetic field is so weak that at most one photon is present at each moment. They had something like 0.5 photons per pulse in average.
It works because a single photon is described by a probabilistic wave function whose behavior completely mimics the behavior of the classical electromagnetic field in these interference experiments. So everything will work even if the "rate of photons per second" is diluted to a tiny number. This disclaimer is analogous to the observation that the interference pattern in a double-slit experiment exists even if we send the photons "one by one" – so the interference pattern can't possibly have anything to do with the interactions of many photons. However, to reconstruct the whole image "UR" in their setup, one still needs many photons that will paint the image as their probabilistic distribution. One photon wouldn't be enough to refresh the whole image.
"Slow light" is a necessary prerequisite and context in which they did their work – a method to undo the "harmful" time evolution of the photon in the dispersive slow-light medium. But the slow light mediums themselves were invented by my ex-colleague Lene Hau in 1999, over 7 years before the paper we are discussing here.
Linguists may summarize the mechanism as: Tomáš Zato Žeseptáš Wolle. Apologies to readers who don't understand it, they may easily ignore this part of the answer.