The Nepal flood video with 7 million views was AI-generated, and here's how to check the next one

The single most-shared piece of “Nepal flood” footage on the internet did not happen. Posted on 27 August with the caption “no one imagined this would happen,” it collected more than 7 million views, 4,100 reposts and 15,000 reactions before the detection service Hive Moderation rated it 97.2 per cent likely to be AI-generated — a verdict since carried by BBC Verify and by X’s own community notes. If you shared it, you were not careless. It was built to be shared.

We have an interest in this. We run a news desk about Nepal, and every photograph we published from this flood was credited to whoever took it, with a link back to the original post so you can check us. That is not a boast — it is the minimum, and the reason we can tell you which of the following is which.

What was circulating, and what it actually was

The three checks that catch nearly all of it

None of this needs software. It needs about ninety seconds.

1. Reverse-search a still. Screenshot one frame and drop it into Google Lens or TinEye. Recycled footage — the Alaska glacier, the Bharatpur Buddha, the Uttarakhand tunnel — collapses immediately, because the older, more-indexed original outranks the repost. This single step catches most misattributed real video, which is the commonest category by far.

2. Check the account, not the clip. Open the poster’s profile and scroll back. An account that posted nothing about Nepal before 26 August and now posts nothing else is farming attention. An account whose earlier uploads are also spectacular disasters from everywhere on earth is a content mill.

3. Watch the edges of the frame. Generative video is confident in the centre and incoherent at the margins. Look at signage and number plates — text is where these models still fail, producing letterforms that are almost a script. Watch for people whose legs change length between steps, water that flows without displacing anything, and buildings that dissolve rather than break. Real debris flows behave like wet concrete; a generated one behaves like smoke.

Detection scores are a useful fourth opinion and a bad first one. Hive and Google’s SynthID are genuinely good, but they return probabilities, and a confident number attached to a wrong answer is how the next round of this spreads.

Why this is not a harmless internet problem

Nepal’s Press Council has opened a complaints portal and ordered takedowns, which tells you the volume. But the damage is not abstract. Fact-checkers documented a woman pleading publicly for help finding her missing husband, and being mocked in her own replies. Families of the missing are reading these comments while they wait.

Recycled footage also wastes rescue attention. When a clip of a different disaster circulates as live, someone has to establish that it is not, and that someone is usually the same short-staffed agency trying to reach a tunnel. Chetana Kunwar of Nepal Fact Check puts the mechanism plainly: AI-generated content and manipulated visuals “make it significantly easier to create false impressions.”

What this means for you

Is any of the flood footage real?

A great deal of it. The security-camera footage from the Gyirong border post, the drone views of Devighat and the Rasuwagadhi tunnel photographs are all genuine, and we have published them with credits and links. The problem is not that everything is fake — it is that the fakes travel faster, because they are engineered to.

Why would anyone fabricate a disaster video?

Attention pays. A clip with seven million views earns platform revenue, sells a page, or seeds a scam donation link. Fact-checkers tracking this flood found fraudulent appeals riding the same wave — which is the practical reason to give only through a route you have verified, and we set those out in where to actually send money.

How do I check whether my own trek is affected?

Ask which river valley it sits in, then check that valley’s status. Our running assessment is on the live updates page, and the honest autumn position — including what we have paused and what we have not — is in should you still book Nepal this autumn.