The devastating flash floods that have swept through Nepal this week have left the country in a state of shock and grief. What began as unusually heavy monsoon rainfall quickly turned into a nightmare, washing away homes, roads, and bridges, and burying entire communities under mud and debris. By Friday, August 28, the official death toll had climbed to 538, and rescue teams were still pulling bodies from the wreckage. For every number counted, there is a family torn apart, a village erased, and a lifetime of memories swept away. The scale of the tragedy is difficult to comprehend, and as people across the world watch from afar, many feel a deep urge to help, to understand, to do something. In the chaos that follows any disaster, social media becomes a lifeline — a place where survivors share desperate pleas, where volunteers coordinate relief efforts, and where news anchors and ordinary citizens pass along updates. But it has also become a breeding ground for something far less helpful: wildfire misinformation. Even as the waters were still rising, videos began circulating online claiming to show the worst of the flooding. Some were genuine, captured by terrified residents on their phones. But others were entirely fabricated, generated by artificial intelligence. These fake clips spread across Facebook, X, TikTok, and messaging apps, often racking up millions of views before anyone could stop and ask a simple question: Is this even real?
The problem, of course, is not limited to Nepal. Every major event that captures the world’s attention now seems to attract its own flock of AI-generated videos. When a war breaks out, when a celebrity dies, when a storm makes landfall, viral footage appears within minutes — and an increasing share of it is fake. The technology behind this is maturing at an astonishing pace. What once looked like crude, glitchy animations now appears smooth, detailed, and disturbingly believable. Modern AI models can simulate realistic scenes with just a short text prompt or a handful of reference images. They can create footage of crowds running through streets, rivers bursting their banks, buildings collapsing, and victims crying out for help. The advancement is genuinely impressive from a technical standpoint, but it is deeply dangerous in the context of a humanitarian crisis. During a disaster, people are scared, anxious, and desperate for information. They are also more likely to share emotionally charged content without pausing to verify it. A video that looks even remotely plausible can be enough to trigger panic, shift attention away from real needs, and even distort the global response. Worse, these fabricated clips can be used to spread politics, sow distrust, or raise money for fake charities. The motivation behind them varies — some creators are simply testing technology, others are chasing clicks and ad revenue, and a small but dangerous subset are deliberately trying to manipulate public opinion. Regardless of intent, the effect is the same: the truth becomes muddier, and real victims are robbed of the attention and aid they desperately need.
So how can an ordinary person, scrolling through their feed in a moment of grief and confusion, tell the difference between a real video and an AI-generated illusion? The first clue is often hidden in plain sight: the length of the clip. Most free AI video generators currently produce footage that lasts only around fifteen seconds, and many are shorter still. If you come across a dramatic clip that lasts only ten seconds or less, it is worth pausing and thinking about whether it might be synthetic. There are exceptions, of course. Plenty of genuine phone videos taken in emergencies are also very short — a frightened person may grab their phone, record a few seconds of chaos, and then run for safety. Length alone is not proof that a video is fake, but it is a useful starting point. The second clue lies in the details within the footage itself. The human eye is not always quick enough to catch these, but if you slow the video down or watch it several times, inconsistencies start to appear. Look closely at the objects in the frame. Do they suddenly change shape, vanish, or alter their color from one scene to the next? Pay attention to lighting and shadows. AI often struggles to replicate the natural behavior of sunlight and reflection, so shadows may fall in the wrong direction, or light sources may behave unpredictably. Reflections on water and glass can be especially revealing, because they are notoriously difficult for AI to simulate with accuracy. Look at faces and bodies as well. Are fingers the right shape? Do mouths move naturally when people speak? Are eye movements smooth and human-like, or slightly jerky and mechanical? Warped facial features, unnatural gaits, and eyes that seem to blink awkwardly are all telltale signs of a generated clip. The more extreme the emotional content, the more carefully you should inspect the visual details — because the purpose of these videos is often to overwhelm your rational mind with emotion.
Another powerful way to expose an AI-generated video is to run a reverse image search on its keyframes. This sounds technical, but it is actually quite accessible to anyone with a basic familiarity with the internet. Essentially, you take a screenshot from the video and feed it into a search engine that looks for matching or similar images elsewhere on the web. If the same scene appears in older posts, unrelated contexts, or stock footage libraries, you will quickly see that the clip was not originally recorded at time of the disaster. Free tools and browser extensions such as InVID or WeVerify are specifically designed for journalists and fact-checkers, but anyone can use them. These tools can break a video down into individual frames and then run those frames through multiple search engines at once, saving a great deal of time and effort. When examining a video that claims to show a specific event like the Nepal floods, this method can be decisive. A real clip of that flood should be entirely new, appearing only in recent posts tied to the event. An AI-generated clip, on the other hand, may have no exact matches online, but it may be part of a larger pattern — perhaps the same creator has published many other synthetic videos, or the same fake footage has appeared in different versions under different captions. Alongside this technical check, it is also wise to look at the broader information landscape. Ask yourself whether reputable news organizations — names you actually trust — have independently confirmed the scene in the video. If a video genuinely shows an important moment from a major disaster, professional journalists and wire agencies will almost certainly pick it up, credit it, and verify it. That does not mean that every real video gets immediate coverage. Breaking news is messy, and verification takes time. Sometimes a genuine video sits unpublished for hours while reporters check its authenticity and track down the original source. So while the absence of news coverage is a reason for caution, it is not by itself evidence that a video is fake. Instead, treat it as a signal to wait and watch rather than share.
For those who want to go even deeper, there are AI detection tools that offer automated video analysis. These are not perfect — the same technology that creates the fakes is constantly evolving, and detection tools sometimes fail to keep pace. But they can serve as an additional layer of scrutiny, especially when combined with manual observation. Some detectors look for subtle digital fingerprints left behind by neural networks, while others analyze statistical patterns in pixels and compression. If an AI detector flags a video as likely generated, it is worth taking that seriously, even if the tool does not offer a definite verdict. Another often overlooked but highly effective step is to check the metadata of the video itself. On a computer, you can often download a video and right-click on the file, then select “Properties” to see the details. Genuine footage recorded on a smartphone or camera will frequently include metadata such as the device model, the camera settings, the resolution, the date and time of capture, and even the GPS location in some cases. AI-generated videos, in contrast, may have no such metadata at all, or they may contain metadata tied to the software used to generate them — information that points not to a phone in Nepal, but to a rendering program in someone’s living room. However, it is important to remember that metadata is fragile. Many social media platforms strip it out automatically when you upload a video, and even when you download a shared clip, it may arrive with no metadata attached. A video without metadata is not automatically fake; the absence of metadata simply means that you have lost one line of evidence. The same goes for screenshots of metadata comparisons available online — these can be misleading because they are easy to fabricate with basic editing tools. Metadata should be treated as one piece of the puzzle, not the whole picture.
Finally, one of the simplest and most reliable checks does not involve technology at all: it involves looking carefully at the source. Before you share a video, take a moment to examine the account that posted it. How long has the account existed? Was it created only a few days ago? Does the profile look complete, or is it missing a photo, a biography, and any history of ordinary activity? What is the posting behavior? An account that pushes out a constant stream of sensational content, particularly videos with alarming captions and emotional appeals, deserves your suspicion. This is especially true if every post seems designed to go viral, to trigger outrage, or to direct attention toward a particular political agenda. Sometimes you will find a pattern: the same account posted dubious videos during previous disasters, or the same clip appears across a network of accounts that all follow each other and share each other’s content. In such cases, you are not dealing with a single misguided uploader; you are dealing with a coordinated misinformation operation. Look at the comments too, but do not trust them entirely. Bots often leave comments designed to make a fake video seem genuine — phrases like “God protect Nepal” or “Why isn’t the media covering this?” are easy to automate. The goal of all these manipulations is to move you quickly, to push you from seeing to sharing before you have time to think. In the end, the best defense is not a single tool or technique, but a slower, more deliberate approach to consuming media during a crisis. Do not let a headline or a dramatic frame steal your compassion and override your judgment. Take a breath. Check the source. Study the details. Search the internet. Ask around. It is not about being cynical; it is about being responsible. The victims of the floods in Nepal deserve the world’s attention, but they also deserve the truth. A shared fake video, however well intentioned, does not help them — it only adds to the noise. What helps is clear information, accurate reporting, and steadfast support for the real work of rescue and rebuilding. By pausing to verify what we see, we honor the people at the center of this tragedy, and we make it far harder for misinformation to flourish. In our hyperconnected world, that small act of care can make an enormous difference.