thisimagedoesnotexist.com

What is This Image Does Not Exist?

This Image Does Not Exist is a browser-based quiz that asks you to judge whether an image was made by a human or generated by a computer. You look at each image, pick "real" or "AI," and the site tells you whether you were right. The page is a JavaScript app, so it needs scripting enabled to run; there are no accounts or downloads described.

How it is useful

• Testing your eye: A quick way to see how well you spot AI-generated faces and scenes, which is harder than most people expect.
• Teaching media literacy: Useful in a classroom or workshop to show that "looks real" is not the same as "is real."
• A conversation starter: Good for demonstrating to others how convincing synthetic images have become.

Who it suits

Casual users curious about their own judgment, teachers covering digital literacy, and journalists or researchers who want a light illustration of the detection problem. It is a game, not a verification tool: a correct guess in a quiz does not mean you can reliably authenticate a real image in the wild.

A practical next step

Try a round, note which cues you relied on (eyes, hands, background text, lighting), then check whether those cues actually predicted your errors. That gap is the useful lesson: human intuition catches some tells but not all, so for anything consequential, verify the source of an image rather than judging it by appearance alone.

How can I tell if an image was made by a human or a computer?

The most reliable way to tell is to look for consistency: a human-made image is internally coherent because one person made deliberate choices, while a computer-generated image often breaks down in small details that a human artist would never get wrong. This Image Does Not Exist turns this into a guessing game, showing you images and asking you to judge whether a person or a machine made them — a useful way to train your eye on real examples.

What to inspect

• Hands and fingers. Count them, check the joints and nails. Extra, merged or oddly bent fingers remain a common giveaway.
• Text and symbols. Signs, labels and logos in generated images often contain garbled or meaningless letterforms.
• Symmetry and repetition. Look at ears, eyes, buttons, windows and background objects: duplicated or mismatched pairs are suspicious.
• Fine structure. Hair strands, fabric weave, teeth and jewelry often blur, melt together or change texture mid-object.
• Physics and lighting. Shadows should fall consistently from one light direction; reflections, glass and water should behave plausibly.
• Background logic. Generated scenes often have vague, melting background details, impossible architecture or objects that merge into each other.
• Overall "feel." An unnaturally smooth, over-polished or slightly surreal quality is a weak signal on its own, so treat it as a prompt to look closer rather than proof.

A practical routine

1. Zoom in and scan the image in a grid, corner to corner, rather than taking it in whole.
2. Check the most failure-prone areas first: hands, faces at an angle, text, and the edges where objects meet.
3. Ask whether every element makes sense in the scene — a plausible object in an implausible place is a clue.
4. Compare against known-human work in the same style. If you are judging a portrait, look at real portrait photography for reference.

Where the judgment is uncertain

These checks are heuristics, not proof. Heavy retouching, filters, AI upscaling and image compression can mimic the artifacts of generation, and modern generators increasingly avoid the classic errors. For high-stakes decisions — verifying a news photo, checking a document or assessing a profile — treat visual inspection as one input and seek the original source, metadata or a reverse image search. The guessing exercise on This Image Does Not Exist is best used to calibrate your instincts, not to settle a dispute.

Why do I keep guessing wrong on the image test?

You keep guessing wrong because the test is designed to defeat exactly the shortcuts most people rely on. This Image Does Not Exist shows you an image and asks whether a human or a computer made it, and the images are chosen to sit in the ambiguous middle: faces that are almost right, textures that are almost plausible. When the only cue is "does this look off?", you are guessing at the edge of your own perception, and near-50% accuracy is the expected result rather than a personal failure.

What usually trips people up

• You are looking for obvious errors. Early AI images failed on hands, ears, text and jewelry. Modern generators often get the gross features right, so the giveaway has moved to smaller things: the way hair meets skin, the symmetry of a reflection, whether background objects resolve into anything.
• You are judging style, not construction. A painterly or heavily filtered photo can look "AI-ish" while being real, and a clean studio-style AI image can look "photographic." Style is a weak signal; physical consistency is a stronger one.
• You are answering too fast. A quick glance rewards confidence, not accuracy. Zooming in and scanning systematically usually helps more than a gut feeling.
• You are anchored by the previous image. After a run of "real" answers, people start expecting a fake, and vice versa. Treat each image as an independent question.

A more useful way to look

Pick two or three checks and apply them every time, rather than hunting for whatever seems odd. For example: follow one line of hair or fabric from end to end, look at what is happening in the out-of-focus background, and check whether any text or repeating pattern stays consistent. Decide what your checks would show for a real photo before you look at the image.

What the score actually tells you

A score near chance means your cues are not discriminating, not that you cannot tell images apart in ordinary life. The practical takeaway is about trust: for anything where authenticity matters — a profile photo, a product shot, a news image — a visual gut check is a weak safeguard. Treat the test as calibration for how much weight to put on "it looks real," and lean on provenance (where the image came from, who published it, whether an original exists) instead.

If you want a second kind of practice, Which Face Is Real uses a similar two-image comparison format and lets you study the specific artifacts that separate the two.

How does this site generate the images I'm judging?

This Image Does Not Exist is a quiz-style site: you are shown images and asked to judge whether each one was made by a person or by a computer. The images it presents are AI-generated portraits, produced by a generative model rather than drawn or photographed for the site. The page itself is a JavaScript app, so the visible text gives no further detail about the underlying model or how each image was created.

In practice, the site is a test of your eye, not an image-creation tool. You are not generating anything yourself; you are comparing your judgment against images the site has already produced with a model. The useful next step is to treat it as a calibration exercise: go through a round, note which cues you relied on, then look again at the ones you got wrong to see whether texture, hair edges, background blur or symmetry gave the answer away.

A concrete reader scenario: a designer checking whether their instincts about synthetic faces are still reliable, or a teacher preparing a short media-literacy exercise. Both get more from the site by discussing why an image looked real or fake than from the score alone.

If you want to go further, compare it with other public demonstrations of generated imagery, such as This Person Does Not Exist, which shows a single generated face per visit. That site is better for studying one image closely; the quiz format here is better for testing your judgment across many examples.

Can I use this site to train myself to spot AI-generated images?

Yes—with an important caveat. This Image Does Not Exist is built as a simple human-or-computer image test, so it gives you exactly the kind of repeated, feedback-driven practice that builds recognition. The caveat: the site's visible content is a JavaScript app, so the page evidence doesn't detail how many images, which generators, or how scoring works. Treat it as a drill, not a complete curriculum.

What it trains well

• Fast pattern recognition. Repeated forced choices ("real or generated?") push you to notice telltale artifacts: over-smooth skin, mismatched earrings, garbled text, odd hands, inconsistent lighting and reflections.
• Calibration. You learn where your instincts over- and under-fire—useful, since false confidence is the main failure mode in spotting synthetic media.
• Low-stakes repetition. A quiz format is more engaging than reading a checklist, which matters for building a habit.

Where it falls short

A single quiz can't cover the full space of generators, and image models improve quickly, so artifacts you learn on one generation may vanish in the next. It also won't teach provenance checks (metadata, reverse image search, source verification), which matter more than pixel-level tells for real-world verification.

A practical routine

1. Do a round on the site and note why you chose each answer.
2. When you're wrong, write down the specific cue you missed.
3. Cross-check with real-world skills: inspect metadata, run a reverse image search, and confirm the original source.
4. Revisit periodically, since the artifacts shift as models improve.

If you want broader grounding in detection and provenance, pair the quiz with resources from organizations that publish on synthetic media, such as DeepMedia or WITNESS, which cover verification workflows beyond a single quiz.

Decision criterion: use the site if your goal is quick, repeatable pattern practice; supplement it if you need to verify images in a professional or journalistic context, where source and metadata checks are essential.

What are the signs of an AI-generated image versus a real photo?

AI-generated images most often give themselves away through small inconsistencies rather than obvious flaws: hands with the wrong number of fingers, garbled text on signs or clothing, jewelry or glasses that don't line up, and backgrounds where objects melt into each other. Lighting is another tell—shadows that fall in directions inconsistent with the light source, or reflections that don't match the scene. Real photos, by contrast, tend to have coherent physics even when messy: consistent depth of field, plausible noise and grain, and text that reads correctly even if blurry.

A quick comparison

These are tendencies, not rules—modern generators handle hands and text far better than earlier ones, so treat any single cue as weak evidence and look for clusters.

A practical test

Zoom in on the areas generators struggle with: fingers, ears, background crowds, and any written words. Then check the light: pick one shadow and trace it back to its source. If you want to calibrate your eye, This Image Does Not Exist presents images and asks you to judge whether a human or a computer made them—useful for seeing how often your instincts hold up.