How to Tell if an Image Is AI-Generated: 7 Checks That Actually Work

The fastest way to tell if an image is AI-generated: zoom in on hands, teeth, and any text in the image, check whether lighting and shadows are consistent, look for warped or “melted” details in the background, then confirm your suspicion with a reverse image search or an AI image detector. No single check is definitive, but an image that fails two or three of them is very likely synthetic.

The latest generators, including Nano Banana (Google Gemini), GPT Image (ChatGPT), Midjourney, and Flux, have fixed many of the obvious giveaways from a few years ago, so the old advice (“count the fingers”) is no longer enough on its own. Here are the seven checks that still work, roughly in the order you should try them.

1. Zoom in on hands, teeth, and ears

Hands remain the most common failure point: extra or fused fingers, fingernails that blend into skin, or a thumb bending the wrong way. Teeth often appear as a single uniform strip rather than individual teeth, and earrings frequently don’t match between ears. Generators have improved here, but complex poses like hands holding objects or interlocked fingers still trip them up regularly.

2. Read any text in the image

Signs, labels, logos, and clothing print are a weak spot for diffusion models. Look for letters that are almost-but-not-quite right, gibberish that resembles a real alphabet, or brand names with subtly wrong spelling. Even the newest models that can render clean text often fail on text that’s small, curved, or in the background.

3. Check lighting and shadows for consistency

Real photos have one physics engine: the sun or the light source. AI images often light a face from the left while its shadow falls to the left too, or show reflections in windows and eyes that don’t match the scene. Eyes are especially useful: both pupils should reflect the same light sources in the same positions.

4. Look for warped or “melted” background details

Generators spend their effort on the subject, so the background is where artifacts hide: fences that change pattern mid-run, chairs with too many legs, crowds where faces smear together, or straight lines (door frames, railings, tiles) that bend for no reason. Architectural logic is hard to fake. Follow any long line across the image and see if it stays plausible.

5. Inspect texture: too smooth, too perfect

AI-generated portraits often have a waxy, airbrushed skin texture with no pores, flyaway hairs, or asymmetry. Fabric weave, wood grain, and foliage tend to repeat in unnatural tiles. If a photo looks like a professional studio shot but claims to be a casual phone snap, that mismatch itself is a signal.

6. Check metadata and Content Credentials

Some generators and cameras now attach provenance data. Content Credentials (the C2PA standard backed by Adobe, Microsoft, OpenAI, and Google) can record how an image was created and edited. You can inspect any image for free at contentcredentials.org/verify. Images made with ChatGPT include C2PA manifests, and Google’s Nano Banana images carry an invisible SynthID watermark. The catch: screenshots and social media re-uploads strip metadata, so a missing manifest proves nothing. Only a present one is informative.

7. Run a reverse image search

Google Lens and TinEye can show you where else an image has appeared. If a “breaking news photo” only exists on social media accounts created last week, or a “real person” appears nowhere else on the internet, be skeptical. Reverse search also catches the other kind of fake: real photos presented with a false story.

When your eyes aren’t enough: use a detector

Manual checks catch obvious fakes, but the best generators now produce images that pass casual inspection. AI image detectors analyze statistical patterns left behind by the generation process, signals invisible to the human eye. You can try one right now with our free AI image checker in your browser, no account needed. For unlimited checks, our iPhone app, AI Detector – Image Checker, returns a color-coded 0-100% confidence score in under 3 seconds, using models validated at 90%+ accuracy and updated for the latest generators.

Detectors have limits too. We cover them honestly in Do AI Image Detectors Actually Work? The strongest verdict comes from combining signals: visual checks, provenance data, reverse search, and a detector score together. If you’re on an iPhone, see our step-by-step guide to checking photos for AI on iPhone.