AI-produced video has reached astonishing levels of sophistication, even to the point of causing concern. Our social media feeds are now inundated with seemingly flawless clips, such as impossibly adorable animals leaping on trampolines, accumulating millions of views on platforms like TikTok, Shorts, and Reels.

As AI-generated content integrates effortlessly into our scrolling habits, it becomes increasingly difficult to discern what is real. So, how can you tell if a popular video is generated by AI?

Regrettably, there isn’t a definitive guide for identifying AI-created videos. “Even if I don’t find the artifact, I cannot say for sure that it’s real, and that’s what we want,” remarked Negar Kamali, an AI research scientist at Northwestern University’s Kellogg School of Management, to Mashable Tech Reporter Cecily Mauran last year.

Traditional indicators—distorted faces, misshapen fingers, unnaturally smooth textures—are becoming more challenging to recognize as technology evolves. Temporal discrepancies are being fine-tuned. However, much like those surreal animal clips caught on fictitious doorbell cameras, the truth often lies in the details where the synthetic facade tends to falter.

The emergence of hyper-realistic AI video tools

Part of the difficulty stems from the technology itself. Tools such as OpenAI’s Sora and Google Veo 3 are now capable of producing cinematic clips with intricate camera maneuvers, realistic lighting, and convincing textures. These platforms have transcended mere novelty—they are encroaching upon professional-level filmmaking, blurring the line between human-created footage and AI-generated content. Consequently, recognizing the “tells” in viral AI videos necessitates keener observation and heightened skepticism.

Consider the video above, for instance, showcasing an entire workshop’s worth of bunnies hopping in flawless synchronization. It’s undeniably charming (and simple to create), but it raises significant doubts upon closer examination.

With this in mind, here are the most effective methods for distinguishing AI-generated viral videos.

1. First, examine the context

Numerous AI videos are staged in oddly specific situations—frequently at night, employing onyx-filter night vision. This isn’t merely for “aesthetic.”

Dark filters conveniently mask minor glitches and frame-to-frame irregularities typical of AI footage.

2. Look for absent device markers

If the video purports to originate from a doorbell camera or security system, check for timestamps, brand logos, and interface overlays. A complete lack of these elements is cause for suspicion. However, the presence of such markers doesn’t necessarily guarantee the video’s authenticity.

3. Observe the physics

Motion in the real world adheres to real-world principles. For instance, animals don’t execute perfectly synchronized, repetitive jumps for ten seconds uninterrupted. For example, notice the tip of this whale, which literally pulls a worker into the ship’s deck.

4. Be aware of the runtime

Shorter videos provide AI with less time to showcase its flaws. That’s why many viral synthetic videos conclude just before something seems “off.”

“If the video is 10 seconds long, be wary. There’s a reason for its brevity,” stated Hany Farid, a UC Berkeley professor of computer science and digital forensics expert, to Mashable.

Similarly, if a longer video consists of very brief clips pieced together, be cautious. Most AI video generators can only produce short snippets. Google Veo 3, the most sophisticated generative AI video model, generates 8-second clips. Sora, from OpenAI, crafts videos ranging from one to 20 seconds long.

5. Listen for audio (or its absence)

Synthetic clips often feature unnaturally clear audio, mismatched background noise, or none at all. “Fabrication coming from them, distorting certain facts…that’s really hard to disprove,” said Aruna Sankaranarayanan, a research assistant at MIT’s Computer Science and Artificial Intelligence Laboratory, to Mashable. Silent or excessively pristine soundscapes can serve as a crucial indicator.

6. Identify AI-text artifacts

AI still has difficulty producing legible writing. Inspect clothing, signage, or packaging within the frame—warped letters, random symbols, or nonsensical text are reliable giveaways. “If the image feels like clickbait, it is clickbait,” Farid told Mashable.

For instance, consider this viral video featuring an emotional support kangaroo. Pay close attention to its vest as the video zooms in.

7. Look for impossible movements

Humans and animals exhibit subtle weight shifts, irregular walking patterns, and micro-movements. AI creations often lack these nuances. Upon close inspection, you might spot strange inconsistencies, such as multiple figures merging into one or vice versa.

“The building added a story, or the car changed colors, things that are physically not possible,” Farid described to Mashable, illustrating temporal inconsistencies.

8. Check for (or the absence of) watermarks

Certain AI video generators—including Sora and Veo 3—automatically embed watermarks or metadata to signal synthetic content. These marks may appear in corners, as faint overlays, or as hidden digital signatures within the file.