AI OR REAL

AI or real? Play the free image quiz

There is no single magic sign that proves an image is AI-generated. The most reliable approach is to compare several visual clues, decide how well the whole image holds together, and avoid overconfidence from one detail.

Learn the method

Try a real question

Which picture was made by AI?

Choose A or B. Get the answer immediately — no account required.

VS

Tap a photo to answer

A repeatable method

Don't guess from one clue. Inspect in layers.

The most useful skill is not memorizing yesterday's AI mistakes. It is learning where several parts of an image must agree with each other.

1. Read the whole image

Start at normal size. Ask whether the scene, expression, lighting and camera feel belong together before hunting for tiny defects.

2. Zoom into high-information areas

Eyes, hair edges, teeth, glasses, fingers, reflections and contact points force many visual relationships to agree at once.

3. Compare, don't stereotype

One smooth cheek or unusual tooth proves very little. Compare several independent clues and look for contradictions.

4. Make a weighted decision

Give more weight to structural inconsistencies than to vague 'AI vibes'. Modern generated images can be extremely convincing.

Visual field guide

Where to look — and why it matters

These are comparison zones, not automatic proof. A strong judgement usually comes from several clues pointing in the same direction.

1

Edges and boundaries

Look where hair, fur, fingers, objects or clothing meet the background. AI can create subtle merging, missing separation or inconsistent fine detail.

2

Texture variation

Real photographs usually contain irregular micro-texture. Generated images can make skin, fabric, food, foliage or surfaces look unusually even or locally inconsistent.

3

Reflections and lighting

Check whether highlights, shadows and reflections agree with each other. Small contradictions can be more informative than the overall realism of the image.

4

Text and geometry

Signs, labels, repeating patterns and rigid shapes are useful comparison points. Modern models are better at them than before, but small structural errors still appear.

Visual lesson · Comparison

Compare concrete facial evidence

Do not judge the overall style first. Compare the exact areas marked on both portraits: eyes, cheek texture and the hair or ear boundary.

The numbered markers correspond directly to the numbered explanations below.

comparison portrait AA123
comparison portrait BB123
1

Eyes and eyelids

Marker 1 sits on the eye area. Compare iris detail, catchlights, eyelid shape and lash edges.

2

Cheek skin texture

Marker 2 sits on the cheek. Compare pore variation, fine lines and whether sharpness changes naturally across the skin.

3

Hair / ear boundary

Marker 3 sits on an outer facial boundary. Inspect how hair, ear, skin and background separate from one another.

Takeaway: A generated portrait can have excellent anatomy. Compare specific corresponding regions instead of relying on a general impression.

Strong clues

Several independent inconsistencies that reinforce each other: geometry, lighting, reflections, anatomy, contact points or local texture that does not fit the same camera and scene.

Weak clues

“Too perfect”, smooth skin, dramatic lighting or one odd feature. Real photography, retouching, compression and portrait mode can create all of these too.

1

A practical way to judge an image

Start with the image as a whole. Ask whether the scene, lighting, perspective and subject all make sense together. Then move inward to small details. This order matters because a single odd-looking feature can exist in a perfectly real photograph, while a generated image may contain several small inconsistencies that only become meaningful together.

Modern image generators are much better than older systems at faces, hands and text, so old rules such as “AI always gets fingers wrong” are no longer dependable. Treat each clue as evidence, not proof. The goal is to build a consistent case from several parts of the image.

  • Check the overall scene before zooming into details.
  • Compare multiple independent clues instead of relying on one artifact.
  • Give extra weight to contradictions in lighting, geometry and object relationships.
  • Remember that compression, blur and editing can make real images look strange too.
2

Why AI detection keeps getting harder

Image generation has improved quickly. Photorealistic skin, natural depth of field, realistic camera noise and plausible backgrounds can now be produced convincingly. That means many obvious artifacts from earlier generations are less common.

At the same time, generated images still have to maintain consistency across thousands of visual relationships. Reflections must match light sources, repeated structures must stay coherent, anatomy must remain plausible, and textures must change naturally across depth and material. Those relationships are where careful comparison still helps.

3

Use context as well as pixels

Visual inspection is only one layer. If the source matters, check where the image came from, whether a reliable publisher provides context, and whether the same image appears elsewhere. Metadata and provenance systems can also help when available.

FaceOrBot focuses on visual recognition because it is a useful skill and a fun challenge, but high-stakes verification should never depend on visual guessing alone.

Remember

A visual clue is evidence — not proof.

A real photo can contain blur, strange teeth, aggressive sharpening, beauty filters or broken-looking edges. An AI image can be almost flawless. Confidence should rise when several independent details fail in a consistent way, not because one feature looks unusual.

Frequently asked questions

Can you always tell if an image is AI-generated?

No. Some generated images are visually indistinguishable from ordinary photographs, especially after resizing or compression. Visual clues can increase confidence, but they are not a guaranteed detector.

Are hands still the best AI clue?

Not by themselves. Hands can still contain errors, but current systems often render them well. Treat hands as one inspection area among many.

Does perfect-looking skin mean an image is AI?

No. Retouching, makeup, lighting and compression can all change skin texture. Compare texture together with lighting, hair, anatomy and background detail.

Can FaceOrBot detect AI automatically?

FaceOrBot is a visual challenge and learning game. It trains your ability to compare real and generated images rather than claiming to provide forensic proof.

Train your eye instead of memorizing rules

FaceOrBot gives immediate feedback after every choice and explains what was worth inspecting. That makes each round a short visual lesson rather than just a score.