Asger Risom
Blurry, short and low-quality clips can currently help reveal AI-generated videos, experts say—but these visual clues may soon disappear as the technology improves.
AI-generated videos are flooding social media feeds, making it increasingly difficult to tell what’s real and what’s not.
While the technology is improving rapidly, experts say there is still one practical clue that can help identify many fakes—for now.
What experts look for
Specialists in digital forensics say one of the first things they assess is overall video quality.
“It’s one of the first things we look at,” said Hany Farid, a professor at the University of California, Berkeley, according to BBC reporting.
This doesn’t mean all AI videos look bad. In fact, many are highly polished. But lower-quality clips can hide subtle flaws that might otherwise expose them.
These can include unnatural textures, small inconsistencies in movement or background elements behaving in unrealistic ways.
Why lower quality can be more convincing
Ironically, poorer video quality can make AI-generated content more believable.
Blurriness, pixelation and compression can mask the small errors AI systems still produce. Experts say creators may deliberately degrade footage to make it harder to scrutinise.
“If I’m trying to fool people… I reduce the resolution… and then I add compression,” Farid explained.
Short length is another common pattern. Many AI clips last just a few seconds, partly because generating longer videos increases the risk of visible mistakes.
A problem that won’t last
These clues may not be reliable for long.
Researchers warn that rapid improvements in AI video tools will soon eliminate many of the visual inconsistencies people rely on today.
“I would anticipate that these visual cues are going to be gone… within two years,” said Matthew Stamm, a professor at Drexel University.
As a result, spotting fake videos based on appearance alone will become increasingly difficult.
The bigger shift
Experts say the long-term solution lies in changing how people evaluate content.
Rather than focusing on visual details, the emphasis will need to shift toward verifying sources, context and authenticity.
“The only thing that matters now is where a piece of content came from… and whether it’s been verified by a trustworthy source,” said digital literacy expert Mike Caulfield.
As AI continues to evolve, the challenge may not be spotting flaws—but learning not to trust appearances at all.
Sources: BBC
