Exploring AI-Generated Imagery: Can You Distinguish Between Nano Banana and ChatGPT Image-2?

In the rapidly evolving landscape of artificial intelligence and image synthesis, discerning the nuances between different AI-generated images has become both an intriguing challenge and a testament to technological advancements. Recently, I conducted an experiment using a highly detailed and specific prompt across two prominent AI image-generation models: Nano Banana and ChatGPT Image-2. The goal was to see if I could identify which image was produced by which system based solely on their visual characteristics.

The Prompt:

To ensure consistency and fairness, I employed an identical prompt for both models. Here is the detailed description provided:

“A hyper-realistic, candid extreme close-up portrait, focusing on one single eye of a man with light blonde hair. The eye is an intense, complex ice blue, with intricate iris patterns, a clean pupil, and an uneven limbal ring, based on the macro perspective of image_3.jpg. The eye shape is average, a natural almond shape. The surrounding skin is highly detailed and unfiltered, showing varied natural pores, a few small moles and freckles, prominent vellus hairs (peach fuzz), and subtle sweat. The skin is not smooth; it has realistic, non-uniform texture with fine micro-wrinkles around the eye. The upper lashes are full and defined, but naturally arranged, without mascara, with individual hairs separate and varied in length. The lower lashes are also defined. Natural, slightly unruly light blonde eyebrow hairs are visible above the eye, with some stray hairs, and a few strands of short, messy blonde hair are visible at the periphery. Lighting is from direct natural daylight filtering through an adjacent window, creating complex reflections and a window-pane catchlight in the pupil. The shot appears candid, resembling a mid-range mobile camera image, with visible digital noise, minor depth-of-field imperfections, a shallow focus plane, and raw, realistic skin and hair textures. Capillaries are visible in the sclera, with minor tears and surface wetness. No beauty filtering or over-sharpening.”

The Results:

After generating two images—each representing the output from Nano Banana and ChatGPT Image-2—I examined their qualities closely.

  • The first image exhibited highly detailed textures, with natural skin irregularities, visible pores, and micro-wrinkles. The lighting and reflection effects closely matched the prompt’s specifications, including a complex window-pane catchlight. The overall realism and presence of subtle imperfections made it look remarkably like a candid human snapshot.

  • The second image, while also realistic, showed slightly less intricate skin textures and a more polished appearance. Some of the fine micro-details, such as individual pore variation and micro-wrinkles, appeared subtly smoother, giving it a less raw and more stylized feel.

Based on these observations and the nuances captured, I identified:

The first image was generated by Nano Banana.

The second image was generated by ChatGPT Image-2.

Conclusion:

This experiment underscores the impressive capabilities of contemporary AI image generators to produce highly realistic and detailed human portraits. While the distinctions can sometimes be subtle, attentive analysis of texture, imperfections, and lighting cues can aid in identifying the source. As these models continue to evolve, the line between AI-generated and real imagery blurs further, challenging us to refine our discernment skills.

Are you able to tell which image is which? Testing such prompts and comparisons can deepen our understanding of AI’s strengths and limitations in visual authenticity.

Leave a Reply

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