Detecting fake images

Generated images have gotten good enough that staring at pixels is no longer a reliable way to catch them. The obvious old giveaways — six-fingered hands, garbled text — still appear sometimes, but the best generators have largely fixed them. So the smartest first move with a suspicious image is not to squint at it, it is to ask about its context and source.

Start with context, because it catches more fakes than any visual tell:

• Where did it first appear, and who posted it? An explosive image from an anonymous account with no history deserves far more doubt than the same image from an established newsroom.
• Is anyone credible showing it? A genuinely dramatic real event is usually covered from many angles by many outlets. A single lonely image that no established source has picked up is a warning sign.
• Does the caption match the picture? A huge share of image misinformation is not generated at all — it is a real photo from another time or place with a false caption. Always separate the picture from the story told about it.

Then, and only as a secondary check, look at the image itself. Modern tells cluster around the things generators still find hard:

• Physics and consistency. Do shadows fall in consistent directions? Do reflections in glasses, mirrors, or water match the scene? Are backgrounds coherent, or do railings, tiles, and patterns bend or merge oddly?
• The boundaries between things. Where hair meets background, where fingers meet an object, where jewelry meets skin — generators often smear or blend these edges.
• Text and fine repeated detail. Small lettering on signs, logos, and labels is still frequently garbled or nonsensical.
• The too-perfect feel. An oddly smooth, flawless, slightly plastic look across skin and surfaces is common in generated images.

Two cautions. First, none of these visual tells is proof — a real photo can have weird shadows, and a fake can be flawless. They only raise or lower suspicion. Second, do not skip the strongest move: the context questions plus a reverse-image search, which is the next lesson. The picture itself is the last thing to trust, not the first.

Recap. With images, lead with context rather than pixels: check who first posted it, whether credible sources are showing it, and whether the caption actually matches the picture — miscaptioned real photos are one of the most common fakes. As a secondary check, look at physics and consistency, the blurry boundaries between objects, garbled small text, and a too-perfect plastic look, but treat every visual tell as a hint, not proof. Then confirm with a reverse-image search, which is the real workhorse and the subject of the next lesson.