Is That Image Real? How to Detect AI-Generated Photos and Videos

A photo goes viral. A video surfaces showing a politician saying something shocking. A product listing uses images that look almost too perfect. How do you know if any of it is real?

AI-generated content has reached a point where it's genuinely difficult to tell the difference with the naked eye. Midjourney, Stable Diffusion, DALL-E, Sora — the tools for creating photorealistic fake images and videos are widely available, easy to use, and improving every month.

The ability to detect AI-generated content has never been more important. And it's exactly what ValidatesAI's new detection feature is built for.

How Big Is The Problem?

The scale of the problem is growing faster than most people realise. Journalists, researchers, HR departments, lawyers and everyday people are increasingly finding themselves needing to verify whether an image or video is real — and until now, there's been no easy way to do it without specialist tools or technical knowledge.

Why Your Eyes Can't Be Trusted

The days of spotting AI images by looking for six-fingered hands or blurry ears are largely over. The latest generation of AI image models produce output that is genuinely indistinguishable from photographs to most human observers — and AI video is catching up fast.

Studies have found that humans correctly identify AI-generated images only around 50% of the time — barely better than a coin flip. Even trained professionals struggle with the latest models. The telling artifacts that used to give away AI images — unnatural skin texture, warped backgrounds, nonsensical text — are increasingly rare in modern output.

You need tools built for this. Which is why we built them into ValidatesAI.

How ValidatesAI's Detection Works

The three sources are combined into a single verdict: Likely AI Generated, Likely Real, or Uncertain. You get a confidence percentage, a breakdown of what each engine found, and Claude's qualitative note where available.

What The Results Look Like

Here's what you might see for three different types of content:

The honest answer is that detection isn't perfect — no tool can give you 100% certainty on every image. Heavily edited real photos can register as AI-generated. Highly sophisticated AI output can slip through. The feature includes a disclaimer noting that results indicate probability, not certainty.

But for the vast majority of cases — a viral social media image, a suspicious product photo, a deepfake video — the combined verdict from three engines is a substantial improvement over your own eyes alone.

Video Detection

Most AI detection tools only handle still images. ValidatesAI's detector also handles video — up to 15MB — using Sightengine's frame-by-frame analysis. It samples frames across the clip and returns an average AI probability score along with the number of frames analysed.

This is particularly relevant given the rise of AI-generated video content on social media — short clips that look real but were generated entirely by models like Sora or Kling.

Who Is This For?

Anyone who needs to verify whether visual content is real. That includes journalists fact-checking images before publication, HR teams reviewing candidate headshots, people who've been sent suspicious images or videos, social media users who want to verify viral content before sharing it, and anyone buying products online where the listing images look suspiciously perfect.

We'll be expanding the detection feature over the coming months — more engines, better video support, and potentially browser extension integration so you can check images directly on any website without leaving the page.

For now, the simplest way to answer the question "is this real?" just got a lot easier.