When I opened it with ImageMagick it appears to be a transparent PNG. Very clever trick, as it seems to use Twitters black background when viewing an image to achieve the effect.
The room is drawn with translucent black pixels that show up on any light background. The stars are drawn with translucent white pixels that show up on any dark background.
That's a different effect. It exploits transparency and the fact that Twitter uses a white background for tweets, but a black background for lightbox images.
The "Ze first image" looks the same whether displayed in Firefox or with the default "Windows Photo Viewer" on Win 8.1. I did have an apple/peach image that did display differently on the same machine about a year ago.
Someone in the twitter thread said that the image looks different on a black background. When I click the image it shows with a white background, so maybe it looks different on mobile? I put the image on a black background manually and here is what it looks like: http://i.imgur.com/91qsoKy.jpg
I didn't either, but interestingly on this image and the ones on the OP, if I tilt my laptop screen (Macbook Air) such that I'm looking at it very obliquely I can see the other image..
Especially "ze first" image as seen in FF and Chrome.
I suppose it is related to how monitors use temporal dithering to fake 24 bit colors when they really have 18 bits per pixel, but why is it more apparent on this kind of picture?
Is it a perceptual issue or a technical one?
It is also apparent when using Windows XP under virtualization. Some windows have checkerboard patterns that flicker.
tl;dr: the liquid crystals in LCDs need to be driven with AC voltage to prevent electrolysis degradation. This is known as "inversion" and there are different patterns used. If the magnitude of the + voltage is different from the - voltage, i.e. the "common" voltage Vcom is not exactly in the middle, the subpixels change brightness slightly on each cycle and this causes flicker. The flickering becomes much more noticeable if the pattern displayed in the image matches the inversion pattern.
Neither of my (nearly 10 years old now... but still working well) LCD monitors flicker on all the images, so this is something that shouldn't be occurring on a good monitor. I suppose manufacturers now are relaxing the tolerances to save cost, justifying it by the fact that flickering is not observable on "most" (debatable) content.
They may be more or less apparent depending on the scaling on the image and the scaling algorithm being applied. Some scaling algorithms apply filters that can make the effect even more weird than normal.
It's not Moiré. If it's displayed 1:1 as the image is designed to be, there won't be any, and even if it isn't and the patterns appear, those patterns shouldn't flicker on a properly adjusted monitor.
Thanks everyone for the details. I don't think it's a moire pattern because the image is displayed at native resolution. So it is a low level calibration issue. Disappointing to see this on a Mac.
I saw someone do something similar in ye olde days of Gravatar. It abused their resizing algorithm to get different images at different scales. It was very simple (and probably couldn't have worked if it was much more complicated) - just the text `i++` on a noisy background. But when scaled it turned into `i--`. In both cases, it was very readable.
It is really an unfortunate over-engineering in the PNG format to include gamma correction. So many problems would have been avoided if PNGs just stored RGB values and let the renderer/OS/driver/screen decide how to display it. Trying to do gamma correction based on various guesses about screens just lead to jarring color inconsistencies.
Here is great article about it. It is a few years old, so some of the problems are fixed now. https://hsivonen.fi/png-gamma/
I also saw it on 4chan first, a long time ago, but I'd bet this trick has been around for many years before that. The gamma of 0.023 also seems to have been the de-facto way to do this.
Here's a good illustration that shows how the pixel values of the 2 images have been "compressed" into two separate ranges:
https://mobile.twitter.com/SciencePorn/status/63454773027106...