How to make AI QR code art that actually scans

Key findings

• 8 of the 14 codes scanned. Every one that scanned did so in under 2 seconds, and the iPhone and the Android phone agreed on every image.
• The scene mattered more than the strength. The cottage garden scanned at every setting, the bright coffee shop only at the highest strength (2.5), and the dark neon street never did.
• Our offline checker recognised only 2 of the 8 codes the phones could read. Scanning with a real phone is the test that counts.
• The image provider’s safety filter blocked 2 of the neon-street images, so no image came back for them.

How we tested

We used three prompts: a bright coffee shop, a neon street at night and a cottage garden. Each one was generated at structure strengths 2.0, 2.2 and 2.5, in a standard and a large code size: 18 combinations. Each scene used one fixed seed, so within a scene only the strength and the size changed.

Every code holds the link to this article, which is 51 characters long. That is longer than a ZapQR short link, and a longer link puts more, smaller squares into the code, so this test is somewhat harder than typical use.

Of the 18 generation requests, 14 returned an image. Two were rejected by the provider’s rate limit, and two neon-street images were blocked by its safety filter.

The 14 images were shown one at a time, in random order and without labels, at 560 × 560 pixels on a computer screen. Each was scanned with the iPhone 14 camera app and an Android phone’s camera from about 30 cm, and a code counted as failed if neither phone read it within 10 seconds.

• Bright coffee shop: “A bright Scandinavian coffee shop counter with latte art, fresh pastries and morning sunlight through large windows, clean white tiles, soft natural light, photorealistic”
• Neon street at night: “A rainy cyberpunk night street with glowing pink and blue neon signs, wet reflections on the pavement, dense city atmosphere, cinematic lighting”
• Cottage garden: “A lush English cottage garden in full bloom with hundreds of tiny colorful flowers, a winding stone path, intricate leaves and petals, highly detailed, photorealistic”

The results

✓ scanned on both phones in under 2 seconds · ✗ neither phone read it within 10 seconds · — no image returned (rate limit or safety filter)

Each image below is the real code we tested. Point your phone at one to try it yourself; the codes that scanned will open this page.

• 

  Bright coffee shop · 2.0 · standard
  Did not scan

• 

  Bright coffee shop · 2.2 · standard
  Did not scan

• 

  Bright coffee shop · 2.2 · large
  Did not scan

• 

  Bright coffee shop · 2.5 · standard
  Scanned on both phones

• 

  Bright coffee shop · 2.5 · large
  Scanned on both phones

• 

  Neon street at night · 2.0 · large
  Did not scan

• 

  Neon street at night · 2.2 · large
  Did not scan

• 

  Neon street at night · 2.5 · standard
  Did not scan

• 

  Cottage garden · 2.0 · standard
  Scanned on both phones

• 

  Cottage garden · 2.0 · large
  Scanned on both phones

• 

  Cottage garden · 2.2 · standard
  Scanned on both phones

• 

  Cottage garden · 2.2 · large
  Scanned on both phones

• 

  Cottage garden · 2.5 · standard
  Scanned on both phones

• 

  Cottage garden · 2.5 · large
  Scanned on both phones

Why the scene matters more than the strength

A phone reads a QR code from the contrast between dark and light squares. The structure strength controls how firmly the AI keeps those squares while it paints the scene.

The cottage garden is full of small light and dark details: petals, leaves and stones. The squares can hide inside that texture, so even the lowest strength was enough.

The coffee shop is mostly white tiles and soft sunlight. Large bright areas wash out the dark squares, so only the highest strength kept enough contrast.

The neon street is the opposite: mostly black, with small bright signs. There is not enough light area for the light squares, and none of its images scanned.

How to make AI QR code art that scans

1. 1

   Choose a scene with mixed light and dark detail

   Gardens, markets, textured materials and busy daytime scenes give the code room to hide. Be careful with scenes that are mostly white or mostly dark.

2. 2

   Keep the link short

   A shorter link means fewer, larger squares. A ZapQR short link is 28 characters and lets you change the destination later.

3. 3

   Start at strength 2.2, and go higher for bright or dark scenes

   In our test a bright scene needed 2.5. If a scene is very dark, change the prompt (for example, daytime or dusk instead of night) rather than relying on strength alone.

4. 4

   Don’t count on code size to rescue a weak scene

   In this test the standard and large sizes gave the same result for every scene where we have both. Choose the size for the look you want.

5. 5

   Scan it with real phones before you print

   Automatic checks missed most of the codes that phones could read. Test with at least one iPhone and one Android phone, then test the printed copy at its final size.

Questions