Section 2.1: Crafting Alt Text

One of my first experiments with ChatGPT Vision involved generating alt text for an image. Alt text is essential for individuals with visual impairments who rely on screen readers while browsing the internet. It describes an image's content, allowing those who cannot view it to grasp its significance.

Creating effective alt text can be challenging and often costly, leading many website owners to neglect this important aspect. To assess ChatGPT Vision's capabilities, I uploaded an image and requested it to produce high-quality alt text.

The result was impressive: “A wall-mounted dispensing system labeled ‘SIGNET’ featuring three translucent containers filled with cleaning solutions. From left to right, the containers are labeled: ‘Hard Surface Sanitizer’ in red, ‘Glass & Multi-Surface Cleaner’ in blue, and ‘Heavy Duty Non-Chlor Washroom Disinfectant’ in green. Above the dispensers are instructional stickers with diagrams indicating proper use. Each container has a measurement scale on its side and is connected to a tubing system for distribution.”

This detailed description not only covered the basics but also provided specific visual elements, allowing visually impaired users to understand the image's content accurately. Remarkably, ChatGPT Vision generated this alt text in about 10 seconds, making it a significant resource for enhancing web accessibility.

Section 2.2: Analyzing Everyday Items

For my next test, I uploaded a photo of my spice rack and asked ChatGPT Vision to identify the spices present and which ones were low on stock.

ChatGPT Vision not only listed the spices but also recognized their brand names:

• Sesame Seeds (Quantity appears good)
• Morton Sea Salt (Cannot determine quantity from the image)
• 365 Ground Ginger (Quantity appears good)
• 365 Cinnamon (Quantity appears good)
• McCormick Chili Powder (Quantity appears good)
• McCormick Coarse Ground Black Pepper (Cannot determine quantity from the image)
• 365 Chili Powder (Quantity appears good)
• McCormick Perfect Pinch Italian (Cannot determine quantity from the image)
• Ground Turmeric (Cannot determine quantity from the image)
• 365 Ground White Pepper (Quantity appears good)
• Ground Cumin (Cannot determine quantity from the image)

Although it struggled with precise quantities, it went a step further by suggesting recipes I could prepare with the spices available.

This illustrates the power of ChatGPT's visual capabilities. It allows for ongoing dialogue about the image, enabling users to ask follow-up questions that expand on the initial responses.

Section 3.1: Identifying Film Types

As a photographer who often works with analog film, I was curious to see if ChatGPT Vision could identify the film type used for one of my photos.

The result was accurate. It identified that the photo was taken on Kodak Tri-X at 400 speed, showcasing the depth of its analytical capabilities.

Section 3.2: Recognizing Dog Breeds

I also uploaded an image of my Bichon Frise, Lance, to see if ChatGPT Vision could determine his breed.

The response stated, “Based on the image provided, your dog appears to have the physical characteristics of a curly-coated breed, potentially a Bichon Frise, Poodle, or a mix of such breeds.” While it hedged its answer by including “Poodle,” which may not sit well with Bichon enthusiasts, it did correctly identify him as a Bichon Frise, emphasizing its ability to analyze features beyond surface-level identification.