Tech & AI

How to Tell If an Image or Video Is AI-Generated (7 Checks That Still Work in 2026)

By Habibur Rahman Sabbir · 2026-11-14 · 7 min read

On this page
How to Tell If an Image or Video Is AI-Generated (7 Checks That Still Work in 2026)

How to Tell If an Image or Video Is AI-Generated (7 Checks That Still Work in 2026)

When you scroll through your feed and spot a picture that looks a little too perfect, your first thought is probably “Is this real?” The short answer is: yes, you can still spot most fakes, but the game has changed. In 2026 the smartest tools are hiding behind sophisticated watermarks and metadata, while the old “count the fingers” trick has lost its edge. Below is a step‑by‑step guide that walks you through the most reliable checks, starting with the source and ending with the visual quirks that still give AI away.

Before you even look at the picture, ask yourself where it first appeared. A quick reverse‑image search can tell you if the file is floating around the web, and whether it’s tied to a reputable source.

  • Google Images: Click the camera icon, upload the image or paste the URL, and hit “Search.” If the same picture shows up on a news site, a stock library, or an older forum post, you’ve got a solid lead.
  • TinEye: This service often catches versions that Google misses. It also shows a timeline of when each instance was first indexed.
  • Yandex: The Russian search engine is surprisingly good at finding AI‑generated art that originates from local platforms.

What you’re looking for is a pattern. If the image first pops up on a site that regularly publishes AI art, or if it appears only in the last few weeks across multiple unrelated accounts, that’s a red flag.

2. Dive into the metadata – Content Credentials

Modern creators are increasingly attaching Content Credentials to their files. These are cryptographic tags that say who made the image, what tool was used, and when it was generated.

  1. Open the file in a viewer that reads EXIF data – on Windows you can right‑click → “Properties” → “Details,” on macOS use “Preview” → “Show Inspector.”
  2. Look for fields like Software (e.g., “Midjourney v5.2”) or a Content‑Credentials block that includes a URL to a verification page.
  3. If the image is from a reputable source, you’ll often see a link to a public ledger (like the Adobe Content Authenticity Initiative) where you can confirm the claim.

When the metadata is missing or has been stripped, that doesn’t automatically mean it’s fake, but it does raise suspicion—especially if the visual quality is unusually high.

3. Look for SynthID or built‑in watermarks

Many AI generators now embed an invisible signature called SynthID. It’s a tiny pattern of pixel variations that only a dedicated detector can read, but some browsers and apps already surface it for everyday users.

  • In Chrome, install the free “AI Detector” extension. When you hover over an image, a small badge appears if a SynthID is detected.
  • On iOS, the Photos app now has a “Source” field that says “AI‑generated” when it finds a watermark.
  • For videos, the same principle applies: services like YouTube have started flagging content that carries a SynthID in the frame‑by‑frame metadata.

If the tool reports a match, you’ve got a definitive answer. If not, move on to the visual clues—AI can still slip past the watermark detection sometimes.

4. Light, shadow, and reflection

Even the best generators still wrestle with physics. Look at the direction of the main light source. Does the shadow line up with the object’s shape? Are reflections consistent across glossy surfaces?

For example, a portrait lit from the left should have a soft shadow on the right side of the nose and under the chin. If the shadow appears on both sides, or if the background light doesn’t match the subject’s lighting, that’s a tell‑tale sign of AI stitching.

5. Anatomy and fine details

Human faces and animal bodies are still the hardest thing for AI to get right. Here are a few quick checks:

  • Ears and hairline: Look for mismatched hair strands that cross the ear or an ear that’s slightly too large.
  • Eyes: Real eyes have subtle reflections and a gradient in the iris. AI eyes sometimes have a uniform color or a “glow” that looks unnatural.
  • Fingers and toes: The classic “count the fingers” trick is less reliable now—generators can produce the correct number—but the joints often look too straight, and the spacing can be off.

When you see a hand where the knuckles are perfectly aligned in a straight line, that’s a clue the image was assembled rather than photographed.

6. Text, logos, and brand details

AI still trips up on real‑world typography. Scan any text in the image—on a sign, a screen, or a label. Look for:

  • Letter spacing that’s too even, as if the text was generated from a single font without kerning.
  • Missing diacritics on foreign language signs.
  • Logos that are slightly distorted—often the angles are a few degrees off, or the colors don’t match the official brand palette.

If you spot a coffee shop sign that reads “Starbuks” with a missing “r,” you’ve probably got an AI‑generated scene.

7. Motion artifacts in video

Videos are where AI fakes get the messiest. Even with the latest diffusion models, the frame‑to‑frame consistency can slip.

  1. Play the clip at a slower speed (most players let you adjust playback speed). Notice any flickering in textures—like a shirt pattern that disappears for a frame and reappears.
  2. Watch for “ghosting” where a background object lags behind the foreground movement.
  3. Check the audio sync. AI‑generated videos sometimes have a slight mismatch between lip movements and speech, especially in low‑budget productions.

When the motion looks smooth but the lighting on the subject shifts oddly from one second to the next, that’s a classic AI artifact.

Why the old tricks are losing steam

Counting fingers used to be a go‑to move because early generators often missed a digit or duplicated one. Today’s models—like Stable Diffusion 3.0 and DALL‑E 4—have been trained on billions of images, so they nail the basic anatomy most of the time. That’s why you’ll see a full set of ten fingers more often, but the joints may still be too perfect, or the nails might lack the subtle variance you’d see on a real hand.

Similarly, looking for “blurry edges” used to work when AI struggled with high‑resolution output. Modern generators now output 8K images with crisp edges, so the blur test is largely obsolete.

Putting it all together

When you suspect something is AI‑generated, start with the source: run a reverse‑image search, check the metadata, and see if a SynthID watermark pops up. If those steps come up empty, move on to the visual checks—light, anatomy, text, and motion. The more boxes you can tick, the more confident you’ll feel about your answer.

At the end of the day, the goal isn’t to become a forensic expert overnight. It’s just to have a handful of reliable tools you can pull out the next time a picture looks a little too polished. With these seven checks in your toolbox, you’ll be able to answer the question “how to tell if image is AI generated” without breaking a sweat.

Comments

No comments yet — be the first to share your experience.

Leave a comment

Comments are moderated — your comment will appear once it has been approved.

Keep reading