← Back to Orai

How to Spot AI-Generated Images

12 proven detection techniques — from symmetry and hands to lighting and text

AI image generators like Midjourney, DALL-E 3, and Stable Diffusion are now capable of creating photorealistic images that can fool almost anyone at a glance. But they're not perfect. Every AI-generated image leaves subtle artifacts — patterns and flaws that reveal its synthetic origin.

This guide covers 12 categories of AI image tells, drawn from hundreds of examples analyzed in Orai's daily puzzles. Master these, and you'll dramatically improve your ability to tell AI from real.

The best way to learn: Don't just read — practice. Play Orai's daily puzzle to test these techniques on real images with instant feedback.

Quick Navigation

  1. Unnatural Symmetry
  2. Hands & Fingers
  3. Text & Signage
  4. Lighting Inconsistencies
  5. Texture Repetition
  6. Geometry & Perspective
  7. Eye Reflections
  8. Hair Detail
  9. Color Saturation
  10. Background Artifacts
  11. Edge Crispness
  12. Perspective & Scale

1. Unnatural Symmetry Common

AI image generators have a strong bias toward perfect bilateral symmetry. Real-world scenes are rarely this symmetric. If an image feels too perfectly balanced — mirror-like, with both sides matching — that's a red flag.

What to look for:

Perfect mirror-balance in compositions that shouldn't be symmetric
Symmetrical arrangements of natural objects (rocks, trees, clouds)
Bilateral symmetry in faces that feels "too perfect"

Pro tip: Real photos have asymmetrical quirks — a tilted branch, an off-center shadow. AI tends to smooth these out into algorithmic perfection.

2. Hands & Fingers Very Common

Hands remain one of AI's biggest weaknesses. Despite improvements in recent models, AI still struggles with correct finger counts, joint anatomy, and hand positioning.

What to look for:

Wrong number of fingers (6+ or 4-)
Fingers blending or fusing into each other
Anatomically impossible joint positions
Hands that are the wrong size relative to the face/body

The finger test: When you see hands in an image, count the fingers and check if each joint makes anatomical sense. This simple check catches a huge number of AI images.

3. Text & Signage Very Reliable

AI image generators don't truly understand language. They learn visual patterns of what text looks like, but can't reliably produce real words. This makes background text one of the most reliable AI detection signals.

What to look for:

Signs, labels, or book covers with gibberish or fake letters
Text that looks correct at a glance but is nonsense up close
Characters from non-existent alphabets
Inconsistent spelling within the same word or sign

Why this works: AI treats text as visual texture, not language. A real "STOP" sign always says STOP. An AI "STOP" sign might say "SOTP" or "ST0P" or something that resembles STOP but isn't.

4. Lighting Inconsistencies Subtle

Real photographs have physically consistent lighting — shadows fall away from the light source, highlights align with it, and light bounces follow the laws of physics. AI often gets this subtly wrong.

What to look for:

Shadows falling in contradictory directions
Highlights that don't align with any visible light source
Shadows that are too soft and perfect (real light is messy)
Missing ambient occlusion in corners and crevices

Pro tip: Pick one obvious light source (the sun, a window, a lamp) and trace whether all shadows and highlights are consistent with it.

5. Texture Repetition Common

AI generates textures (brick walls, paving, fabric, grass) by repeating patterns. Real surfaces have random wear, variation, and imperfection. AI textures can look like seamless wallpaper — too regular, too clean.

What to look for:

Brickwork or paving that's suspiciously uniform
Fabric or grass patterns that repeat identically
Surfaces that are too clean — no stains, cracks, or wear

6. Geometry & Architecture Reliable

AI struggles with geometric precision. Lines that should converge (like railroad tracks or building edges) often don't. Vanishing points are misplaced. Angles are slightly off in ways that feel "impossible" once you notice them.

What to look for:

Architectural lines that should converge but don't
Vanishing points in wrong positions
Arches or curves that are geometrically impossible
Wall-ceiling-floor junctions that don't connect properly

7. Eye Reflections Portraits

In real photos, eyes reflect the environment. If someone is outdoors, you'll see sky or trees reflected. AI eyes often look flat, glowing, or have catchlights (the white reflection of a light source) that don't match any visible light.

What to look for:

Catchlights that don't match the scene's lighting
Perfectly symmetrical studio catchlights in a non-studio setting
Flat, glowing eyes with no environmental reflection
Slightly mismatched pupils between left and right eye

8. Hair Detail Portraits

Real hair has flyaways, clumps, varied texture, and follows the laws of physics. AI hair often looks painted on — individual strands are too uniform, hair falls in impossible directions, and there's a telltale halo effect where hair meets skin.

What to look for:

Hair that's too perfect — no flyaways or clumps
Individual strands that are suspiciously uniform (like spaghetti)
A halo or blur effect where hair meets the forehead/face
Hair physics that defy gravity or wind direction

9. Color Saturation Common

AI tends to pump up color vibrancy. Skies are impossibly blue, grass is neon green, sunsets are extra dramatic. Real cameras capture flatter, more muted tones with natural color casts.

What to look for:

Over-saturated colors that look like HDR
An "everything glows" quality
Color grading that's too perfect and uniform
Skies without natural gradient variation

10. Background Artifacts Very Common

AI is good at rendering the main subject but often loses detail in the background. Objects in the distance dissolve into abstraction, and background elements don't make sense when you trace them.

What to look for:

Backgrounds blurred into mush behind a sharp subject
Distant objects that dissolve into abstraction when traced
Background elements that don't connect logically
"Trees" or "buildings" in the distance that aren't real trees or buildings

11. Edge Crispness Subtle

Real camera lenses produce natural softness at edges (slight chromatic aberration, lens blur). AI renders edges with unnatural digital crispness. Objects have clean, sharp outlines that no real lens would produce.

What to look for:

Edges that are too sharp, too clean
A halo around the subject (compositing artifact)
Object boundaries that are unnaturally precise
Missing chromatic aberration at high-contrast edges

12. Perspective & Scale Subtle

AI struggles with wide-angle perspective distortion. Distant objects sometimes appear too large or too sharp. The horizon line may bend subtly. The scale of objects relative to their distance can be inconsistent.

What to look for:

Distant objects that are too large or too sharp
A subtly bent or curved horizon line
Scale inconsistencies between near and far objects

Putting It All Together

No single technique catches every AI image. The key is to combine multiple checks:

  1. First impression: Does it feel "uncanny"? Trust your gut.
  2. Quick checks: Hands, text, symmetry — these are fast and catch most AI.
  3. Deep checks: Lighting, geometry, edges — for images that pass the quick checks.
  4. Context check: Does the image make logical sense? Are backgrounds consistent?

The more you practice, the faster these checks become automatic. That's exactly what Orai is designed for — daily practice with instant feedback.

Ready to put these techniques into practice?

Play Today's Puzzle →

Frequently Asked Questions

How can I tell if an image is AI-generated?

Check for unnatural symmetry, incorrect hands and fingers, garbled background text, inconsistent lighting, texture repetition, impossible geometry, wrong eye reflections, painted-looking hair, over-saturated colors, background artifacts, and unnaturally crisp edges. Combine multiple checks for the best results.

What is the most reliable AI image detection technique?

Background text is one of the most reliable signals — AI garbles letters in signs and labels. Hands and fingers are also highly reliable, as AI still struggles with correct anatomy. Lighting consistency is a strong tell for more sophisticated AI images.

Can AI generate photorealistic images?

Yes. Modern AI generators can create images that are extremely difficult to distinguish from real photos. However, they still leave subtle artifacts. As AI improves, detection becomes harder — which makes daily practice and building visual literacy more important than ever.

What is the "finger test" for AI images?

The finger test is simple: count the fingers in any image showing hands. AI frequently produces wrong finger counts (more or fewer than 5), fused fingers, or anatomically impossible joints. It's one of the fastest and most reliable AI detection checks.

How do I practice spotting AI images?

Play Orai — a free daily game with 6 images. Each guess gives instant feedback and a specific detection tip drawn from the 12 categories above. Just 2 minutes a day builds lasting visual literacy.

Why does AI mess up text in images?

AI image generators learn visual patterns, not language. They produce text that looks like real writing but is actually meaningless. A real "EXIT" sign says EXIT. An AI "EXIT" sign might say "EX1T" or "EXET" or contain fake characters.

Orai — orai.fun — Spot the AI. Daily. Practice AI image detection, learn how to spot AI-generated photos, and build your visual literacy with a free daily game.

Read the how to play guide, check our FAQ, or learn about Orai.