The arrival of AI video generators such as Google’s Veo and Kling AI, with the ability to produce minutes of photorealistic footage from a single text prompt, has made it harder to distinguish between synthetic content and authentic, real-world footage.

In the future, these tools are likely to become even more capable as several AI research labs and startups are building models that can simulate realistic motion, lighting, and physics.

The AI industry has been taking steps toward watermarking and provenance by adopting standards like C2PA Content Credentials and Google’s SynthID that essentially attach verifiable origin information to media. But on the other side of the equation, detection is still in its early stages.

Story continues below this ad

Hence, the burden currently is on every user to verify something before they share it on social media. Today, spotting an AI-generated or AI-processed video has become a digital literacy skill that every social media user needs to learn. Here are the biggest red flags that a video is AI-modified, and what to look for.

Facial expressions and body movements

AI-processed videos featuring photorealistic people generally show them blinking too fast or too slow. Their expressions also seem oddly regular and do not exactly match the emotions of the scene. Additionally, the texture of their skin, eyes, and hair may shift from one frame to another.

Body movements are also common giveaways as AI-generated, photorealistic people may move too smoothly, without the imperfect corrections a real human body tends to make.

Unsynced audio and lip movements

Another common sign of AI-manipulated or synthetic video is that lip movements do not exactly line up with the words being spoken, especially toward the end of a sentence.

Story continues below this ad

However, voice cloning has progressed by leaps and bounds in the past two years. Hence mismatched audio and lip movements may not be the most reliable indicator.

Unreal lighting, shadows, and reflections

AI video generators are known to struggle with capturing real-world lighting in line with physical rules. For instance, a shadow might point the wrong way, a reflection might show something slightly or completely different from the subject, and the lighting might shift illogically between cuts.

Platform labels and disclosures

As mentioned above, several major social media platforms such as YouTube and Instagram have adopted visible watermarking and labelling standards. This means that AI-generated content on these platforms are likely to be visible somewhere in the corners of the frame.

These watermarks are applied using a combination of creator disclosure and automated detection. On Meta-owned Instagram, for example, look for a small “AI info” or “Altered or synthetic content” tag near the video.

Story continues below this ad

However, plenty of synthetic content often slips through unlabeled. There have also been instances of real-world footage being incorrectly labelled as AI-generated in the past.

Trace the video to the source account

Users who want to go a step further can scrutinise the account that posted the video for details such as when the account was created, its posting history, username and profile picture, etc. Generally, accounts that have been newly created and post at high volume should be treated with some skepticism.