AI-generated images are everywhere now, and they're getting harder to spot with every model release. Deutsche Welle published a fact-check guide breaking down the telltale signs that an image came from Midjourney, DALL-E, Stable Diffusion, or their ilk—and it's required reading for anyone who doesn't want to get fooled.

Anatomy of an AI-Generated Image

The most reliable giveaways tend to appear in hands, teeth, text, and backgrounds. AI models still struggle with rendering fingers correctly—expect extra digits, missing knuckles, or weirdly fused hands. Text within images is notoriously mangled, with gibberish characters where words should be. Backgrounds often look slightly smeared or contain impossible architecture that your brain registers as 'off' without immediately understanding why.

Texture and Lighting Inconsistencies

Look for skin textures that are too smooth or pores that appear in patterns that don't match real human biology. Hair strands frequently blend together unnaturally, and lighting sources within an image often contradict each other—shadows falling different directions on the same subject, highlights that don't correspond to any visible light source.

Reverse Image Search Is Your Friend

When in doubt, run suspicious images through Google Images or TinEye. Many AI-generated images circulating online are slightly modified versions of existing photos, and reverse search can reveal the original. Newer tools like Hive Moderation and Getty's AI detection service offer more sophisticated analysis if you need to verify authenticity professionally.

The Metadata Problem

Most legitimate news organizations strip metadata from published images, which ironically makes it harder to trace origin—but AI-generated images often lack EXIF data entirely or contain generation parameters that expose their synthetic origins. Check the file properties before assuming an image is authentic.

Key Takeaways

  • Examine hands, teeth, and text carefully—AI models still struggle here
  • Look for lighting inconsistencies and impossible shadows
  • Use reverse image search to trace potential origins
  • Check metadata and EXIF data when available
  • When uncertain, treat images as potentially synthetic until verified

The Bottom Line

We're past the point where 'I saw it online' means anything. If you're building any kind of content moderation pipeline or just want to maintain your sanity, bookmark this DW guide and share it with everyone who still forwards you AI-generated nonsense pretending to be real photos.