The Old Checklist Is No Longer Enough
Three years ago, spotting a synthetic photo was often a matter of counting fingers or squinting at background text. Six-fingered hands, garbled signs, teeth that merged into a smooth smear—these were reliable giveaways. That era is over. The old rules no longer hold the same weight.
Current image generators like Midjourney v6, Stable Diffusion XL, and DALL-E 3 have become remarkably adept at the very details that used to betray them. Hands are usually correct. Text is often readable. Backgrounds are coherent. The result is that simple visual inspection has become unreliable for many categories of synthetic images, especially photorealistic portraits. The gap between what looks real and what is real has nearly closed. Researchers at Northwestern University found that most people perform only slightly better than chance when asked to distinguish synthetic faces from real ones.
Visual Clues That Still Sometimes Work
These checks are not universal. A clean result on all of them doesn’t confirm a photo is real. But they’re worth examining, especially if something already feels slightly off. If you spot one of these, it’s a strong hint—but not a guarantee.
- Hands: Count the fingers and study knuckle proportions. Hands gripping an object—a cup, a phone—are particularly difficult to render correctly; the contact points between fingers and the object often look wrong.
- Teeth: When the subject smiles, look for teeth that are too uniform in size, too white, or too perfectly aligned. Real smiles have variation, and synthetic ones sometimes blend at the edges or produce too many teeth for a normal mouth.
- Ears: The inner ear is a complex three-dimensional shape. On a real photo, it has genuine detail. Synthetic ears often look smooth or slightly flattened, especially toward the back edge.
- Hair near edges: Where hair meets the background, synthetic strands often have an unnaturally smooth transition, especially against a complex backdrop.
- Jewelry and accessories: Rings, earrings, and necklaces tend to blur slightly into the skin or have an inconsistent metallic quality. These items are generated as part of the image rather than as separate objects, which shows at close range.
- Background logic: Look at signs with readable text, bookshelves with identifiable titles, clocks with correct time. Real photos have those. Synthetic backgrounds tend to have plausible-looking but incoherent details when examined closely.
- Skin texture: Real skin, even well-lit skin, has pores, fine lines, and texture variation. Synthetic skin is often too smooth—not obviously fake, but unusually clean, more like a rendering than a photograph. This is subtle and doesn’t apply to all generators.
- Eyes: Pupils should be consistent in size, and light reflections in both eyes should match. Iris texture should have natural variation. Synthetic eyes are usually good now, but inconsistencies between the left and right eye still appear occasionally.
A photo can pass all of these checks and still be synthetic. These are simply the things that are still sometimes wrong. For a deeper visual guide on how to tell if a photo is AI, see this resource.
What Detection Tools See That You Cannot
Synthetic images are produced through a fundamentally different process than photographs. A camera captures light hitting a sensor. A generator samples from a probability distribution learned from millions of training images. The output images look similar to photographs, but the underlying data has different statistical properties. That difference is what detection models exploit.
Detection models are trained to identify these differences. They examine patterns in noise, color gradients, and pixel-level frequency distributions that differ between photographed and generated images. A clean, well-rendered synthetic image can be undetectable by eye but still identifiable by a model analyzing the raw image data.
This is why detection tools catch things visual inspection misses. The image looks real, but the math underneath it looks different from a real photograph. The limitation is that generators keep improving. A detection model trained on older outputs may not catch newer ones as reliably. The best tools retrain on new generator outputs regularly, so accuracy figures need to be interpreted with the specific tool and timeframe in mind. A verdict is a probability, not a proof.
A Practical Workflow for Suspicious Images
Start with a quick visual pass using the checklist above. Look at hands, teeth, ears, hair edges, jewelry, background logic, skin texture, and eyes. This first pass helps you catch poorly rendered images, of which there are still plenty. It won’t catch everything, but it’s a useful filter.
If the image passes visual inspection but you still have doubts—especially for a dating profile, a news photo, or a job applicant’s headshot—use a detection tool as a second step. For example, Faux Spy checks any image visible in Chrome. You right-click the photo (or hover for an Investigate button), click it, and you get a verdict with a confidence score in a few seconds. The five detection categories—generated photo, real photo, generated art, digital art, illustrated—cover the main types of generated and real images you’ll encounter. The tool offers three checks per day free, with no account required.
The tool’s limitation is important: accuracy varies by generator and timeframe. No detection tool is perfect. Use it as one piece of evidence, not a final verdict. Combine it with your visual assessment and any context you have.
Different Image Types, Different Red Flags
Not all synthetic images are created equal. Photorealistic portraits of human faces are the hardest to spot visually and the most important to check with a tool. They are used in fake social media profiles, catfishing, and fake reviews. These are the images most often used to deceive.
Synthetic scenes and environments—landscapes, city streets, indoor spaces without people—often have background logic issues. Signs, storefronts, and text on surfaces are usually the first place to look. Text that looks almost right but not quite is a common red flag.
Synthetic art—illustrated, painterly, or fantasy-style images—is less likely to deceive someone into thinking it’s a real photo, but it matters in art theft and disclosure contexts. Artists and commissioners increasingly use detection tools to verify whether work was hand-made or generated.
Enhanced real photos are a separate category: the base image is real but has been significantly altered. Detection of this type is harder and less reliable than detecting fully generated images.
A final note: visual inspection is a first pass. For anything where accuracy matters, use a detection tool as the second step. And remember that a photo can pass all visual checks and still be synthetic. The goal is not perfect certainty but a reasonable, layered assessment.
Brooklyn-born astrophotographer currently broadcasting from a solar-powered cabin in Patagonia. Rye dissects everything from exoplanet discoveries and blockchain art markets to backcountry coffee science—delivering each piece with the cadence of a late-night FM host. Between deadlines he treks glacier fields with a homemade radio telescope strapped to his backpack, samples regional folk guitars for ambient soundscapes, and keeps a running spreadsheet that ranks meteor showers by emotional impact. His mantra: “The universe is open-source—so share your pull requests.”