Generated AI has improved at an incredible pace until many of us struggle to distinguish whether we are talking to humans and AI chatbots. AI-generated images and videos have also emerged dramatically over the past few years, with the best AI photos and videos being hardly distinguishable from the real thing.
Mostly, but not perfect.
How to find videos generated in AI
Some people want to avoid using AI generations from text to video, but from YouTube to Tiktok, AI-generated videos have recently emerged.
Why is this important? Videos are an important modern source of information, so to avoid misinformation, you need to understand what is realistic and what is not.
There are still some big giveaways to keep in mind if the video is artificial intelligence or not. None of these are absolute, and as AI continues to improve, some will become more reliable (as we'll see in the final section!). But for now there are still dead giveaways to look for when trying to detect AI use.
1. Impossible physics
One of the important giveaways that the video you are watching was created by AI is a sign of behavior that violates the laws of physics. Certainly, special effects means that not all legal videos always adhere to those laws, but they can usually convey the difference between intentional special effects and unintended anomalies that are incorrectly included.
Examples of impossible physics include objects that change courses without external forces being applied, people who are either too high or appear to belong to the matrix, liquids that behave like solids (or vice versa). These problems arise in videos generated by AI, due to artificial intelligence models that rely on data rather than actually understanding real physics.
2. Bad transitions between scenes
If you've seen examples of videos generated by bad AI, you're probably familiar with the bad transitions between scenes. However, these sample videos usually highlight comically ridiculous or entertaining transitions, but even more subtle videos can provide clues. Find people or objects that morph into something else. Something that is impossible in real life.
Clear examples of bad transitions include pointless cuts between scenes. If humans are filming something, they make sure the cuts have some narrative meaning, but that's not a concern with generative AI models. This is due to the way AI generates video frame-by-frame, and no logic applies unless the human prompts are detailed enough to follow the story.
3. Human movement/expression
Human representations are difficult to replicate and are one of the best things to convey when checking whether a video was created using AI. We can always look at each other and see the expressions being formed in people's faces, so we know the subtle clues to look for. Generic artificial intelligence is unable to develop skills to replicate human expression and emotions, as it does not yet fully understand the nuances associated with human expression and emotions.
In AI-generated videos, these issues reveal themselves in obvious and subtle ways. Movement is the most obvious way, as human arms and legs are shaking. Although less obvious, still notable examples are the convulsive mouth, the unnatural blinking of the eyes, and the expression that suddenly moves from the extreme to the other.
4. Visual background noise
Videos shot by real people capture the entire scene within frames, but AI often overlooks the background and instead focuses on getting the subject right. However, it provides an easy way to detect whether you have created a video using AI. Look beyond the subject of the foreground and see what's going on in the background. It may be nonsense, but it could also contain blurry textures, tears, or artifacts.
People who use AI to generate videos are very focused on getting the subject right, but don't care much about what's going on in the background. Text-to-video AI models do their best to remind you of the right background, but they struggle to get rid of visual noise. Therefore, you may see flickering textures and wobbling trees. And these bright signs can occur in some frames, but not in others.
5. Discontinuation of behavior and emotions
If the main subject of the video you are watching is human, look for discrepancies between what is being done (or either of them) and the emotions they are displaying. Often, the two are subtly turned off, and sometimes very off. As an example, if someone is at risk and has a casual, goofy smile on their face, it's a complete mismatch.
AI handles the creation of people and objects with wide brush strokes, but when they try to show someone's lips moving or playing across their face, they often leave a desire. This is because AI struggles with small details. Nailing human emotions requires great caution for AI models that rely on the data they supply, not on a natural understanding of nuances and meaning.
6. Nonsense Sequence
Last but not least, there is a nonsense sequence in the list of dead giveaways when determining whether the video you are watching is AI or not. Again, these issues appear in many of the poor AI video examples that are widely shared on social media. Nonsense sequences are stories with zero meaning. Ask why and how in the video and it's more likely to be AI.
A classic example of nonsense sequence is when Smith eats spaghetti. Smith may eat spaghetti in real life, but as they are usually shown in these AI videos, he rarely becomes a gorge on it (although the difference between early AI videos and the latest models like the Veo 3 is surprisingly different). Another example I've seen many times is cooking in the kitchen as Gordon Ramsay cooks, sheds light on everything around him, and does a few other things that are totally and completely nonsense.
Overall: Trust your instincts
All of the above should be looking for specific things, but the biggest thing when trying to detect video generated by AI is to trust your instincts. Humans have great instincts as long as they choose to use them. Just as you should not believe everything you read online, you should not accept that a video appears to be. Because, thanks to some great AI video generators and detailed prompts from skilled prompt engineers, they are increasingly being generated by AI.
The increase in AI is unfortunately consistent with a decrease in attention span between the general population. So we fear that many people don't pay sufficient attention to what they see to detect AI generation. This is a big problem. Therefore, pay attention to the video you are watching and trust the instincts you will need to do so. And always remember that if it looks too good it is probably.
