You can't just glance at a photo anymore. You have to look at the details.
In an era when photographs of events that never occurred spread across the internet faster than truth can follow, the ability to read an image critically has become a form of civic literacy. Experts in artificial intelligence are offering practical guidance to help ordinary people distinguish the fabricated from the real, as the gap between synthetic and authentic imagery narrows to near invisibility. The urgency has reached legislative chambers in Chile, where lawmakers are beginning to reckon with the social weight of synthetic media. What was once a technical curiosity has become a question about how we know anything at all.
- AI-generated images now achieve a realism so convincing that even attentive viewers can be deceived, and false photographs routinely travel thousands of shares before any correction reaches them.
- The spread of synthetic disaster scenes, fabricated protests, and impossible events is no longer a hypothetical threat — it is a daily feature of online life with real consequences for public trust and political reality.
- Experts are equipping the public with a practical visual vocabulary: malformed hands, impossibly perfect faces, incoherent backgrounds, contradictory shadows, and the conspicuous absence of any verifiable source.
- Chile's Congress has moved the issue from the feed to the floor, drafting legislation to regulate how AI may use human faces, voices, and likenesses — a signal that governments are beginning to treat synthetic media as a structural problem.
- The window for detection remains open, but narrowing — the tells are still there for those trained to look, and the question now is whether visual literacy can scale faster than the technology erasing its own traces.
Every day, the internet fills with photographs of things that never happened — fires, earthquakes, protests that exist only in pixels. People see them, believe them, and pass them along thousands of times before anyone asks whether the image is real. The problem has grown serious enough to reach Chile's Congress, where lawmakers are drafting rules to limit how artificial intelligence can use human faces, voices, and likenesses.
Two years ago, spotting a fake was easier. The technology left obvious scars. But AI has evolved faster than our ability to see it, according to Tomás Vera, an AI specialist and director of Zenta Group. "You have to look at the details," he explains. Misinformation spreads in the time it takes to hit share, and by the time fact-checkers catch up, the false image has already traveled further than the truth.
Vera and other experts have identified a set of visual signals that can help. The most persistent is the hands — fingers fused together, bent at impossible angles, or simply the wrong number. The second is almost the opposite: faces that are too perfect, with flawless skin, unnatural symmetry, and a vacancy in the eyes that real faces never carry.
Backgrounds often betray what the subject conceals. Text becomes gibberish. Objects warp or repeat. Shadows point in conflicting directions and reflections appear where they shouldn't. These environmental inconsistencies accumulate like evidence against the image's credibility.
But perhaps the simplest test requires no technical knowledge: ask where the image came from. Many false photographs circulate with vague language and no traceable source — no news organization, no photographer's name, only anonymous accounts and forwarded chains. A real photograph has a history. When an image exists only in the viral moment, that absence is itself a kind of answer.
The stakes are not abstract. False disaster images trigger panic. Synthetic photographs of public figures shape elections. The technology will only improve. But for now, the tells remain — small imperfections in the seams between the artificial and the plausible, visible to anyone who learns to look before they share.
Every day, the internet fills with photographs that never happened. Fires that didn't burn. Earthquakes that never shook the ground. Protests that exist only in pixels. People see them, believe them, and pass them along—thousands of shares before anyone stops to ask if the image is real. The problem has become urgent enough that it's reached the floor of Chile's Congress, where lawmakers are drafting rules to limit how artificial intelligence can use faces, voices, and likenesses. But for most of us scrolling through our feeds, the question remains: how do you know?
Two years ago, spotting a fake was easier. The technology left obvious scars—warped proportions, impossible geometries, the digital equivalent of a forged signature. But AI has evolved at a pace that outstrips our ability to see it. The realism now reaches levels that can fool almost anyone, according to Tomás Vera, an AI specialist and director of Zenta Group. "The technology improved so much that you can't just glance at a photo anymore," Vera explains. "You have to look at the details." This shift from obvious error to subtle deception is what makes the current moment so precarious. Misinformation spreads in the time it takes to hit share, and by the time fact-checkers catch up, the false image has already traveled further than the truth ever will.
Vera and other experts have begun laying out a practical guide for the rest of us—a set of visual signals that can help separate the real from the fabricated. The first and most persistent tell is the hands. Fingers remain one of the hardest things for AI systems to render correctly, even as the technology advances. Look for digits that are fused together, bent at angles that no human joint allows, or simply the wrong number of them. A hand with six fingers or a thumb that bends backward is a red flag that should stop you before you share.
The second signal is almost the opposite problem: faces that are too perfect. AI-generated portraits often display skin without blemish, symmetry that nature rarely produces, and a kind of emptiness in the eyes. Real faces carry the texture of living—asymmetries, the small marks of time, the particular light that reflects in an actual person's gaze. When a photograph looks like it belongs in a movie rather than in the world, that's worth questioning.
The background often betrays what the subject conceals. Text becomes unreadable gibberish. Objects warp or repeat. Windows tilt at angles that defy physics. A person might appear twice in the same scene. These environmental inconsistencies are harder for AI to manage than the central figure, and they accumulate like evidence. Similarly, light behaves strangely in synthetic images. Shadows point in conflicting directions. Reflections appear where they shouldn't. The illumination feels flat or impossible, as if multiple light sources were fighting each other.
But perhaps the simplest test requires no technical knowledge at all: ask where the image came from. Vera points out that many false photographs circulate with vague language—"they say this just happened," "this is happening now"—but no actual source. No news organization published it first. No photographer's name appears. It lives only in anonymous accounts and forwarded message chains. A real photograph has a history. It was taken by someone, published somewhere, verified by people whose reputation depends on accuracy. When an image appears with none of that scaffolding, when it exists only in the viral moment, that absence itself is a kind of answer.
The stakes are not abstract. False images of disasters can trigger panic. Synthetic photographs of public figures can shape elections. Deepfakes of ordinary people can destroy reputations. The technology will only improve, making detection harder. But for now, at least, there are still tells—small imperfections in the seams where the artificial meets the plausible. The question is whether we'll learn to see them before we share.
Notable Quotes
The technology improved so much that you can't just glance at a photo anymore. You have to look at the details.— Tomás Vera, AI specialist and director of Zenta Group
One of the simplest ways to detect fake content is to ask yourself who originally published that image. If it only appears in anonymous accounts or message chains with no backing from credible media, there's reason to be suspicious.— Tomás Vera
The Hearth Conversation Another angle on the story
Why do hands keep failing? It seems like such a specific problem.
Hands are geometrically complex—twenty-seven bones, countless joints, infinite possible positions. When AI generates an image, it has to understand not just what a hand looks like, but how it moves in three-dimensional space. Most of the time it gets it wrong.
So if I see a photo that looks perfect except the hands are weird, I should assume it's fake?
Not assume—suspect. It's one signal among several. But yes, hands are reliable enough that experts still point to them first.
What about the faces that look too perfect? Doesn't that seem backwards? Shouldn't we trust something that looks clean and clear?
That's the trap. Real faces are messy. They have pores, asymmetries, the small imperfections that make someone look alive. When something is flawless, it often means no one was actually there when the photo was taken.
And the background—why is that easier to mess up than the person?
Because the AI is trained to focus on the subject. The background is secondary, almost an afterthought. It's where the system's attention fails.
So the real test is just asking: where did this come from?
Yes. If no one credible published it first, if it only exists in forwarded messages and anonymous accounts, that's usually the answer right there.