ImageGuesser
M04. SURFACES

Texture & Text

AI is excellent at local texture and terrible at global rules. It knows what one brick looks like, but not how a thousand bricks fit together—and it knows what letters look like, but not how to spell.

The Gibberish Giveaway

Generators were trained on pixels, not dictionaries. They learned what text looks like—high-contrast shapes arranged in lines—without ever learning to read. The result is "pseudo-text": signage that resembles language at a glance but dissolves into an alien alphabet when you actually read it.

Read the Background, Not the Subject

Modern models can spell a word you explicitly prompt for ("a sign that says STOP"). What they still fail at is incidental text: store names, book spines, bottle labels, badges. Ignore the subject entirely and read everything else in the frame.

  • Merging Letters: An 'H' that flows into an 'E', or strokes that don't belong to any alphabet.
  • Mid-Word Decay: A word that starts correctly and dissolves into squiggles halfway through.
  • Font Drift: Style or weight changing between letters of the same word—"dream logic" typography.

Broken Patterns

Architects and tailors rely on math: a skyscraper has identical windows, a plaid shirt has lines that connect. AI treats a pattern as a vibe rather than a rule. It starts drawing on the left, and by the time it reaches the right, it has effectively forgotten the count and alignment.

Architecture

Follow one row of windows or tiles across the whole image. AI window grids often start square at the bottom and turn wavy, crooked, or merged near the top. Stairs and fences that garble into ramps are the same failure.

Fabric

Check plaid and floral prints at folds and seams. Real patterns follow the geometry of the fabric; AI patterns drift, vanish, or turn to noise where the cloth bends—and plaid lines rarely match across a button placket.

Texture Blending

In a photograph, a scarf and a neck are separate objects. In a generated image, everything is painted from the same digital "soup," so boundaries between materials often smear. The tell is a painterly transition where one material becomes another.

  • Hair-Clothing Fusion: Long hair resting on a shoulder that dissolves into the fabric pattern instead of lying on top of it.
  • Embedded Jewelry: Necklaces and earrings that look surgically implanted in the skin rather than resting on it.
  • Vanishing Frames: Rimless glasses whose hardware melts into the eyebrows or the bridge of the nose.

Always inspect the points where two objects touch or overlap—that's where the illusion is thinnest.

Accessory Asymmetry

People match their earrings. AI doesn't. It generates the left ear, then effectively forgets what it did by the time it renders the right one. This short-term memory problem applies to anything that comes in pairs or should connect across the image.

  • Mismatched Earrings: A pearl stud on one side, a gold hoop (or nothing) on the other.
  • Broken Headphones: A band that goes over the head but never connects to the ear cup.
  • Uneven Frames: Glasses where one lens is square and the other is round, or a collar whose points don't align.

The "AI Look"

Beyond individual glitches, generated images share a default aesthetic. Models are trained to prefer "pleasing" images, so they drift toward movie-poster lighting, centered composition, and perfect background blur—even when the prompt called for a casual snapshot.

Real Photos Are Messy

A genuine phone snapshot has awkward framing, harsh mixed lighting, and clutter. If a "candid" image has golden-hour rim lighting, cinematic color grading, and flawless bokeh, treat that perfection itself as evidence. This is a probabilistic signal—professional photography is also polished—but combined with any other artifact from this course, it should push your confidence hard toward "generated."

Common Manifestations

The Coffee Shop Sign

A realistic cafe photo where the menu board reads "COFEEEEE" or "EXPRS&O" with symbols that don't exist in any alphabet.

The Melting Skyscraper

A glass tower whose window grid starts square at street level and turns into wavy lines or merged blobs near the top.

The Skin-Lace

A lace collar whose pattern literally becomes veins or skin texture where it meets the neck.

The Mismatched Bling

A portrait with a diamond stud in the left ear and a dangling gold hoop in the right.

Field Checklist

Before you guess, run through these checks:

  • 1
    Read every piece of background text: signs, labels, book spines, badges. Gibberish ends the debate.
  • 2
    Follow one row of windows, tiles, or fence posts across the entire image and watch for drift.
  • 3
    Check that clothing patterns follow the folds of the fabric and match across seams.
  • 4
    Inspect every point where two objects touch: hair on shoulders, jewelry on skin, glasses on noses.
  • 5
    Compare left and right accessories—earrings, lenses, collar points—for symmetry.
  • 6
    Ask whether the image is suspiciously beautiful for its supposed context.