Everyone checks the fingers, and AI mostly fixed the fingers. These are the checks that still work, starting with the two a photographer trusts most.
That's the rule professional fact-checkers work by, and it's the one thing to hold onto from this whole page. One odd detail can have an innocent explanation. Two or three flags together mean slow down before you share.
The checks below are ordered the way I'd run them: the visual ones first, because they take seconds and your eye is better at this than you think.
Find the shadows. In a real photo, every shadow points away from the same light source. AI loves to light the face from one side and the background from another.
The photographer's note. Light direction is the first thing a photographer reads in any image, and it's the thing AI still fakes worst. If the scene feels "off" and you can't say why, it's almost always the light disagreeing with itself.
A real lens blurs gradually: the further from the focus point, the softer things get. AI gives you the airbrushed look, everything evenly smooth, or everything impossibly sharp at once.
The photographer's note. Depth of field is physics, and physics is consistent. When a face, the hair, and the mountains behind it are all equally crisp, no camera did that.
Read every sign, label, logo, and document in the image, letter by letter. Misspellings, strange capitalization, letters that melt into each other.
Why it works. Image tools paint letters instead of typing them. One wrong letter in a logo you know settles the question on its own.
Parallel lines that bend. Railings that don't meet the stairs. A person in the background the wrong size for where they're standing.
Why it works. Find two things that should naturally line up and check if they do. Real geometry always agrees with itself; generated geometry often almost does.
Ask 4 quick questions: Does this event seem realistic? Would news outlets be covering it? Is the image built to trigger a strong emotion? Does the caption match what's visible?
Why it works. Images made to fool you are made to move you. A big emotional pull with zero coverage anywhere is its own red flag, before you even study a pixel.
Look at the account that posted it. Pages pumping out dramatic images every hour are usually farming clicks, and platforms now add AI labels to flagged content on Instagram, Facebook, and YouTube. Check for the label before you check anything else.
Drop the image into Google Images or TinEye. A real news photo shows up in multiple reputable places. A generated one usually traces back to a single social post and nowhere else.
Tools like SightEngine and Hive can scan an image, and Google stamps its own AI images with an invisible SynthID watermark. Treat every detector score as evidence, never proof: they miss edited or compressed images, and they mislabel real photos too. The stacked checks above still make the call.
AI video has different tells than AI stills, so it gets its own checklist. Find me on social and comment REAL on any of my videos, and I'll send it to you.