AI sharpening distorts Kraków attack images
TLDR
- Demagog says widely shared "sharpened" faces of the Kraków suspect carry a SynthID watermark showing Google AI made them.1
- Komputer Świat says AI upscaling guesses missing pixels, and turned small text in one image into garbled letters.2
- PAP says SynthID flagged several suspect-face images, and that a patch was added to a social media photo.3
Polish fact-checkers and PAP reported on 8 November that images circulating after the 7 November shooting in Kraków's Main Market Square, which Polish media report killed six people, were produced by running blurry frames through AI "sharpening" tools.13 Demagog says the results are invented detail, not evidence, and that sharing them can point suspicion at innocent people.1
Why it matters: upscaling a blurry frame is a tempting step in a fast-moving verification, but the checkers describe it as generating a plausible face, not recovering one. Detection tools and watermark checks gave them a way to show it.123
Driving the news:
- Watermark checks: Demagog says the most widely shared "face of the attacker" image was made by trying to sharpen blurry images with AI and carries a SynthID watermark, which it says shows a Google AI model produced it. It adds that other users produced a different face from the same material, and that the pattern on the suspect's T-shirt changed when the AI raised the resolution.1
- Symbols: Demagog says an image attempting to explain symbols on the suspect's sweatshirt and helmet carries a SynthID watermark indicating an OpenAI model, and that a Hive AI analysis gave further confirmation. It says an image showing a Ukrainian flag on the suspect's vest also carries an OpenAI-model watermark and came from an unreliable source.1
- Evading detectors: Demagog says some images were shared as screenshots of a monitor, which it suggests may be an attempt to fool detection tools.1
- How upscaling fails: Komputer Świat says models "draw in" the most probable missing pixels. It gives a garbled "pdiclq" in place of "policja" ("police") as an example of small text turning into random characters, and says an image of a police officer kneeling by the suspect is also AI. It says upscaling fails where fidelity matters, such as licence plates or forensic evidence.2
- Official response: Radio Kraków, citing PAP, says Deputy Prime Minister and Digital Affairs Minister Krzysztof Gawkowski posted on X that his ministry met overnight with its disinformation team, and that sharpened images "often mislead" and sharing or commenting on them amplifies disinformation. PAP says its own SynthID check found several suspect-face images were altered with Google AI.3
- Altered real photos: PAP says photos of the suspect taken from social profiles were also altered. In one Facebook photo a patch with three arrows, an antifascist symbol, was added although the suspect did not wear it.3
The big picture: PAP identified five main disinformation narratives. Beyond the images, it lists a claim of a second shooter on St Mary's Church, which it says reached 1 million views on X and which police attribute to a priest stepping onto a terrace, and a claim that the suspect was a Polish soldier, which the Defence Minister denied.3 It also reports a post falsely presented as the suspect's brother, with nearly 2 million views, whose subject called it a "provocation by foreign bloggers". A NASK expert told PAP these narratives were spread in the Russian and Belarusian information space.3 Małopolska police said the first patrol began work 23 seconds after the report, countering claims of a slow response.3
Single-source note: the SynthID, Hive AI and PAP detection results are the checkers' reports and were not reproduced here, and the images were not reviewed. The OpenAI attribution of the SynthID watermark is Demagog's wording. Komputer Świat is dated 7 November, before the window, but covers the same wave of images. The suspect is deliberately not named.