AFP Used UncovAI to Expose a Viral AI Hitler-Face Building
In October 2026, an image of a building with a facade resembling Adolf Hitler's face spread across social media in more than a dozen languages. AFP traced it. The image is AI-generated, and the building does not exist. To prove it, AFP ran the image through UncovAI.
AFP tested four versions of the image with UncovAI. Three scored 100% probability of AI generation.
What is the viral Hitler-face building image?
The image shows a residential building whose windows read as eyes, whose canopy reads as brows, and whose central column and shuttered window read as a nose and moustache. The sloping roof completes the haircut.
Many posts treated it as a joke. Some called it the new headquarters of far-right parties in Slovakia, Germany or Austria. Others shared it as a real photograph. Some gave street addresses in Austria or the Czech Republic, and a few users claimed to have visited. The same picture appeared in posts in Bulgarian, Polish, English, Portuguese, French, Spanish, Latvian, Estonian, Slovenian, Ukrainian, Chinese, Arabic and Persian, with tens of thousands of shares in total, according to AFP.
Even where the tone was satirical, the comments filled with people asking whether the house was real.
How AFP verified the image
AFP could not identify the original creator. The earliest traces it found were TikTok posts from 9 and 10 October 2026, since deleted, and a Latvian post on Threads from 11 October. From there, the team worked through four checks.
- Reverse image search. Wider versions of the picture showed the whole building and its surroundings.
- Visual inconsistencies. A partly cut-off car, a traffic sign with no post, and trees that seem to grow straight out of the building.
- Watermark check. Every tested version came back positive in OpenAI's Verify tool, which reported the content as generated with OpenAI tools.
- AI detection with UncovAI. AFP ran four versions of the image through the UncovAI detector.
One misleading post also carried an overlay with a spelling error, "Avarge house in Germany". Small details like that rarely survive in real photographs.
How AFP used UncovAI to confirm the image
AFP, one of the world's largest news agencies, used the UncovAI image detector as part of its fact-check. It tested four versions of the picture: the one from the verified Slovak post, two wider versions showing the whole building, and the Latvian copy from Threads.
The Slovak-post image and both wider versions scored 100%. The copy shared on Threads scored 11%. AFP published the UncovAI results in its fact-check alongside its other evidence.
AFP's Medialab team notes that format, aspect ratio and compression can shift detection results. Searching for earlier copies is good practice, but it doesn't guarantee a higher score.
That gap is the point. A score is one piece of evidence. Here the other checks, the watermark, the visual errors and the missing building, all agreed with the high scores.
When Google and Meta tools got it wrong
Two automated answers pointed the other way.
- Google's AI Overview. On a reverse image search, it said the building exists at Neklanova 30 in Prague's Vyšehrad district, designed by Josef Chochol in 1913 to 1914. AFP checked Google Maps. The real building at that address is a cubist apartment house, Hodek's tenement, and looks nothing like the viral image. The Prague claim in some posts may trace back to this answer.
- Meta's AI detection tool. In a reply on Threads, it called the picture a real photograph and tied it to a house in Swansea, Wales, which drew press coverage in 2011.
The city of Linz, named in several Bulgarian and German posts, told AFP on 15 October 2026 that the building as shown does not exist there. A police spokesperson said the real Hitler birth house in Braunau, reopened in July 2026 as a police station, does not resemble the photo.
What readers and newsrooms can take from this
- Check the claim, not only the pixels. A named street address can be tested in minutes on a map.
- Test several versions. Wider, earlier and unedited copies give more to work with. Cropped or compressed reposts can score lower.
- Combine methods. Reverse search, visible flaws, watermarks and detectors support each other. None should carry a verdict alone.
- Treat AI answers as leads. An AI summary or a chatbot's "real photo" verdict still needs a source behind it.
- Keep detection close to where you work. The UncovAI browser extension lets you check an image without leaving the page.
Frequently asked questions
Did AFP use UncovAI to verify the image?
Yes. AFP tested four versions of the image with the UncovAI detector. Three scored 100% probability of AI generation and one scored 11%, and AFP combined these results with other checks.
Is the building that looks like Hitler's face real?
No. AFP's investigation found the image was created with artificial intelligence, and no such building exists at any of the addresses claimed in social media posts.
How did AFP show the image was AI-generated?
AFP combined reverse image search, a check for visual inconsistencies such as a cut-off car and trees growing out of the facade, an invisible watermark check, and AI detection tools including UncovAI.
Can AI image detectors be wrong?
Yes. Format, aspect ratio and compression can change a detection score. Here, three versions scored 100% on UncovAI while a copy shared on Threads scored 11%, so results should be read alongside other evidence.
Why did Google and Meta tools say the building was real?
According to AFP, a Google AI Overview placed the building at a real address in Prague, and Meta's detection tool called the image a real photo and linked it to a house in Swansea. Both were wrong.
Sources
- AFP Fakty, fact-check published 30 October 2026, on the AI-generated building image, including its tests with UncovAI. Written by Bozhidar Anguelov and Elena Crisan (AFP Bulgaria, AFP Austria).
Run the same check AFP ran
A convincing picture takes seconds to generate and days to debunk. Test it with UncovAI before you share it.
Get Started Free →