How to Tell If a Photo or Video Is AI-Generated

‍TL;DR

• The old advice for spotting AI images, weird hands, garbled text, too-perfect skin, is increasingly out of date. Current AI image and video tools have fixed most of those obvious tells.
• The most reliable signs left are subtler: inconsistent lighting and shadows, physically impossible details in the background, and a strange, slightly-too-smooth uniformity across the whole image.
• For video specifically, unnatural blinking patterns, audio that doesn't quite sync with mouth movement, and lighting that doesn't shift naturally as the subject moves are still the most useful things to watch for.

Spotting AI-generated images used to be almost easy: 

• Extra fingers 
• Nonsense text on signs 
• An unnervingly smooth, waxy look to skin. 

Current tools have largely fixed those specific mistakes, which means the old checklist doesn't work nearly as well as it used to. Here's what's actually still worth looking for.

Why the old advice stopped working

Early AI image generators struggled with specific, predictable things, hands with the wrong number of fingers, text that dissolved into meaningless squiggles, faces that looked slightly too polished. Those were genuinely reliable tells for a while, and a lot of the advice still circulating online is built entirely around them. The problem is that image generation tools have specifically targeted and largely fixed those exact weaknesses, which means checking for six-fingered hands today catches a shrinking fraction of actual AI-generated images.

What's actually still worth checking for images

Lighting and shadows that don't quite agree with each other

This is one of the more durable tells. AI image generators are good at making any individual object look convincing, but less reliable at making every shadow in a scene consistent with a single light source. A shadow falling the wrong direction, or a reflection that doesn't match what should be visible, is still a meaningful signal.

Background details that don't hold up to scrutiny

The main subject of an AI-generated image is usually the most carefully rendered part. Backgrounds, especially busy ones with lots of repeated elements like crowds, patterns, or architectural details, are where inconsistencies tend to survive. Look for details that repeat oddly, blend into each other, or don't quite make physical sense the longer you look at them.

An overall texture that feels slightly too uniform

Real photography has a certain visual noise and inconsistency to it, skin isn't perfectly even, fabric has genuine texture variation, surfaces catch light unevenly. AI-generated images sometimes carry a subtle, hard-to-articulate smoothness across the whole frame that real photos don't quite have, even when no single element looks obviously wrong.

What's different about spotting AI video

Blinking and micro-expressions

Natural blinking has an irregular rhythm and involves small, involuntary facial movements around it. AI-generated video sometimes produces blinking that's too regular, too infrequent, or isolated from the small muscle movements that normally accompany it.

Audio that's almost, but not quite, in sync

Slight mismatches between lip movement and audio, more noticeable on certain sounds than others, remain one of the more reliable signs in AI-generated or manipulated video, even as the technology improves.

Lighting that doesn't respond naturally to movement

As a real person moves through a space, the lighting on their face and body shifts in ways that are physically consistent with the environment. AI-generated video sometimes fails to update lighting convincingly as the subject moves, leaving it looking subtly static or mismatched to the background.

Why this keeps getting harder

The specific visual mistakes that used to make AI-generated content easy to spot are being fixed one at a time, which means the reliable signals keep shifting toward subtler, harder-to-articulate inconsistencies rather than disappearing entirely.

There's no permanent checklist here, what's a reliable tell today may not be one in a year, as the specific weaknesses being exploited get addressed by newer tools. The most durable approach isn't memorizing a fixed list of signs, it's staying skeptical of a single image or video as proof of something significant, and looking for independent confirmation when it actually matters.

FAQs

Is it still possible to spot AI-generated images by looking at hands? 

Less reliably than before. Early AI image tools frequently rendered hands incorrectly, but this has been largely fixed in current tools, so it's no longer a strong signal on its own.

What's the most reliable sign that a photo might be AI-generated? 

Inconsistent lighting and shadows tend to be one of the more durable signs, since AI tools are better at rendering individual objects convincingly than keeping an entire scene's lighting logically consistent.

How can I tell if a video has been AI-generated or manipulated? 

Look for unnatural blinking patterns, audio that's slightly out of sync with lip movements, and lighting that doesn't shift naturally as the subject moves. These remain some of the more consistent signs even as the technology improves.

Will these detection methods keep working in the future? 

Not indefinitely. As AI tools improve, the specific weaknesses being used to detect them tend to get fixed, meaning reliable signs shift over time rather than staying constant. Healthy skepticism toward unverified images and video matters more than any specific checklist.

‍

‍