Fabricated AI Images of Nepal Flood Devastation Go Viral

Images purporting to show the flooded town of Trishuli Bazaar in Nepal before and after the catastrophic flood of August 26, 2026 turned out to be entirely AI-generated. Fact-checkers at AFP, Lead Stories, and Australia's AAP FactCheck independently confirmed the fabrication, detecting invisible SynthID watermarks from OpenAI's tools embedded in the files.

What the Images Showed

The composite, made up of two frames, showed a river valley first intact, then wrecked by water and mud. The Facebook post's author did not state a location explicitly, but an accompanying comment suggested it depicted the aftermath of the flooding in Nepal. Only in a follow-up post on Instagram did someone specify that the images allegedly documented the destruction in Trishuli Bazaar, a town that was in fact flooded by the real disaster.

Trishuli Bazaar sits on an inner bend of the Trishuli River, in a densely built floodplain. The real destruction there was documented by, among others, Reuters on August 26, 2026. The problem is that the topography in the fabricated images bore no resemblance to the town's actual terrain or its real layout of buildings.

How the Fabrication Was Detected

Lead Stories submitted both photos and their composite to OpenAI Verify, a tool for detecting content generated by the company's models. The system found SynthID watermarks in the files, a technology developed by Google DeepMind to invisibly tag synthetic images, video, and audio. The trace appeared both in the composite image and in each of the two separate "before" and "after" photos.

Further confirmation came from a mismatch with authentic satellite imagery of the region. The buildings in the fabricated images matched neither Trishuli Bazaar's actual layout nor the terrain along the river's bend. Australia's AAP FactCheck independently verified the same case and reached identical conclusions.

I can't believe what I'm seeing is real. It looks almost surreal - from a post by Facebook user @jimmatuga accompanying the fabricated images

The Scale of the Real Disaster

Behind the fake images, a genuine tragedy was unfolding. According to the US Geological Survey (USGS), the flood was caused by the collapse of a glacier on the northern slope of Langtang Lirung peak in the Himalayas, where a block of ice and rock broke off at an altitude of about 5.2 kilometers above sea level and fell nearly 1.2 kilometers straight down into the valley. The event triggered a magnitude 5.2 tremor, recorded by seismometers around the world.

The flood wave traveled at roughly 50 meters per second, reaching settlements as much as 80 meters above the river's normal level. The districts of Rasuwa, Nuwakot, Dhading, Gorkha, and Chitwan in Nepal were affected, along with Gyirong in China's Tibet Autonomous Region, where the Gyirong Port border crossing, among other things, was destroyed. As of September 5-6, confirmed deaths had reached 1,385, with more than 5,405 people still listed as missing.

Why It Matters

The Nepal flood case illustrates a pattern fact-checkers increasingly describe around major disasters: a real tragedy immediately draws fake AI-generated content or footage recycled from older events, and these, fueled by emotional captions, spread faster than official reporting. Alongside the AI images themselves, falsely captioned footage of real disasters from other countries also circulated, as did conspiracy theories linking the flood to alleged involvement by the American HAARP research facility.

For Polish news and social media audiences, this is another example that the availability of free watermark-detection tools like OpenAI Verify is becoming a practical verification resource, though it still requires informed use. The mere fact that an image looks dramatic and fits headlines about a real disaster is no proof of authenticity.