Meta has been touting its auto-generated image descriptions for years now. Most recently, the tech giant launched its newest app, Threads, without an alt text field. It stated in a Meta newsroom post that, "The core accessibility features available on Instagram today, such as screen reader support and AI-generated image descriptions, are also enabled on Threads."
Unfortunately, despite how long Meta's AI-program has been around, it still falls short when it comes to accurately describing images so that they're accessible for folks who use assistive devices to access digital content. The auto-generated image descriptions normally lack key details and important context. They more closely resemble a string of random keywords rather than an image description.
To better illustrate the gap between auto-generated and custom image descriptions, below are ten images and the descriptions automatically generated for them by Facebook as well as the custom descriptions I wrote for them. The difference between the two should be quite obvious.
Auto-Generated Image Description from Facebook: "May be an image of 2 people, child, people swimming, ocean and beach."
Custom Image Description: "Two young women in bikinis splash through the shallow waters of a sandy oceanside beach."
Auto-Generated Image Description from Facebook: "May be an image of parasail."
Custom Image Description: "Upward view of a colorful hot air balloon floating through a cloud-filled blue sky."
Auto-Generated Image Description from Facebook: "No photo description available."
Custom Image Description: "Round platter of 55 raw half-shell oysters arranged around a small ceramic bowl filled with red dipping sauce and three metal spoons."
Auto-Generated Image Description from Facebook: "May be an image of 1 person and hat."
Custom Image Description: "Profile shot of a stylish young woman wearing a patterned silk scarf and a sleeveless blouse that matches the sea green wall behind her. Atop her head is a wide-brimmed, felt-green hat that she delicately grasps with one hand."
Auto-Generated Image Description from Facebook: "May be an image of pool."
Custom Image Description: "Overhead view of a tropical resort's circular pool surrounded by sun-bathing chairs, hot pink beach umbrellas, and lush foliage. A few resort guests swim in the pool and lounge in the chairs."
Auto-Generated Image Description from Facebook: "May be an image of crocodile."
Custom Image Description: "Extreme close-up of a crocodile's green eye. The skin around it is glistening and wet."
Auto-Generated Image Description from Facebook: "May be an image of 1 person and phone."
Custom Image Description: "Young woman in a floppy beach hat and soft green dress taking a picture with her phone in front of large banana leaves."
Auto-Generated Image Description from Facebook: "May be an image of omelet and French toast."
Custom Image Description: "Overhead view of a white table laden with breakfast items like glasses of juice, baskets of toast, fresh fruit, bacon, ham, scrambled eggs, and various pieces of flatware."
Auto-Generated Image Description from Facebook: "May be an image of 1 person and twilight."
Custom Image Description: "Woman joyfully holding up a transparent umbrella lined with twinkling lights. Behind her, the horizon line glows orange from the last light of the setting sun while the dark sky overhead is dotted with constellations and a single shooting star."
Auto-Generated Image Description from Facebook: "May be an image of flower."
Custom Image Description: "Bed of tulip flowers in various shades of pink, orange, red, and yellow. They're not fully open yet and the petals are covered in dew drops."
This post was inspired by a tweet from Scribely.