Can You Tell The Difference Between a Photograph and an AI Generated Image?

Can You Tell the Difference Between a Photograph and an AI-Generated Image?

Artificial intelligence (AI) has rapidly evolved, moving from the domain of specialized industries to accessible technology for everyday use. Today, AI tools are widely available, enabling anyone to experiment with image generation, text-to-image prompts, and other visual effects. But as AI capabilities improve, an intriguing question arises: Can we tell the difference between a photograph and an AI-generated image? To explore this, I set up a simple experiment comparing a photograph of my dog with both a stock image and an AI-generated version of the same breed.

The Experiment: Real vs. AI-Generated

To conduct this comparison, I started with a photograph of my dog; a black, male, 55-pound Portuguese Water Dog. I then searched for a similar image on a free stock photography site, selecting a photo that matched my dog’s appearance as closely as possible. Finally, I used AI to generate an image of a black, male, 55-pound Portuguese Water Dog. I used DALL-E as my AI provider for this task, entering prompts to recreate my dog’s appearance.

The AI’s initial attempt was reasonable, but only depicted the dog’s head. I quickly realized that my initial prompt was too vague and began making adjustments. Here’s a summary of the modifications I applied to make the AI image more accurate and realistic:

  1. Full body view
  2. Less curly and more wavy hair
  3. A more realistic, photographic style
  4. Added white on the upper chest and front paws
  5. Black mouth detail
  6. Straighter hair
  7. Shorter hair on the head and ears
  8. Sitting position instead of lying down
  9. Softer, more wispy hair texture
  10. Proportion adjustments for a smaller head and larger body

Each of these adjustments slightly altered the image, helping it better resemble my dog. However, after multiple tweaks, I encountered a limit with DALL-E’S image generator. An error message informed me that I had reached a request threshold, requiring me to wait before making more adjustments.

The Results: Can You Guess Which Image is Which?

Once I had all three images; the photograph I took, a stock image, and an AI-generated image, I compared them.

Can you tell which image is my photograph, which is a stock photograph, and which is AI?

Have you figured it out?

My photo is the first image, followed by the stock photo (The second image is a stock photograph sourced from Pexels, used under their free download option), and finally, the AI-generated image.

Despite my best efforts, the AI image still looked like a computer-generated image. It required numerous modifications, and even then, it lacked the lifelike qualities of the first two photos. The AI struggled with the subtleties in texture, fur shape, and overall natural appearance that were more evident in the actual photographs.

Conclusion: The Future of AI-Generated Realism

This experiment left me a little disappointed with the AI-generated result. Despite advancements, AI still struggles to capture the full realism of a photograph, especially when generating textures and facial expressions. However, as AI technology continues to advance, it is only a matter of time before these distinctions become increasingly subtle, making it harder to tell the difference between real and AI-generated images.

If you are curious, try comparing images and see if you can tell the difference. After all, AI is just getting started.

Leave a Reply

Your email address will not be published. Required fields are marked *