YouTuber’s pixelation fail: private video part unmasked in hours

One YouTuber laid out a challenge to his viewers: $50 to anyone who could de-pixelate a heavily obscured section of his video, intended to hide private folders. It was cracked in hours, demonstrating the vulnerability of blurring as a privacy measure.

Moving pixelated images contain enough information to restore what’s hidden behind the mask.

In a video testing an external SSD hard drive, YouTuber and software developer Jeff Geerling used heavy pixelation to censor portions of his Mac screen showing an internal folder structure.

He added a short written note: “If anyone can de-pixelate what’s on my network share, I’ll give you $50 ;)”

From the first glimpse, the task seems impossible because the pixels are very large, leaving almost no information. However, viewers quickly noticed that Geerling moved the folder around a bit, changing the mosaic with every move. And that was enough to reveal the text hidden behind.

“FYI, not one but TWO people have successfully de-pixellated the folder listing I censored in the video! I'll try to get the payment to both of them soon, and we'll also see if one or both can publish their method,” the YouTuber pinned a comment.

Developer shared the method

A developer who goes by the alias KoKuToru on GitHub detailed the whole process, which took around four hours. KoKuToru extracted dozens of different frames and the hidden information from them using FFmpeg (software library for handling multimedia files) and GIMP. The developer shared proof of concept on GitHub, as discovered by TechSpot.

“I was more or less jumping around like a small child, couldn't believe it worked that good. Seeing it with my own eyes was like magic,” KoKuToru shared.

“Saw the video, looked doable. Did it.”

The developer analyzed moving pixelation patterns as a moving “fence,” with some gaps revealing parts of the content behind them. While the window was moving, the “pixelated” contents changed differently than the moving window.

It took two iterations for KoKuToru to develop and fine-tune the fill algorithm that reverse-engineered the image. The approach was similar to upscaling, where a low-resolution image is made into higher resolution using data extracted from motion.

The final output revealed readable text – folder names – and content, astonishing even the creator.

The code was openly shared on GitHub and has already garnered significant interest: it has been starred 340 times, and 35 forks already exist.

The project highlights vulnerabilities in basic pixelation as a privacy measure. In the past, various pixelation and blurring patterns have also been demonstrated to be insecure and can be undone. The Depix tool uses matching patterns against a known font database to de-pixelate even still images. De-blurring has also been automated using AI tools.

Using black bars or other irreversible redactions is recommended to hide sensitive content.

Geerling, in a later video, admitted: “It's scary how fast someone can de-pixelate or de-censor video footage – provided it has enough data within it for a neural network to process.”