Midjourney V5 rolls out with new features promising accessibility

"Give it a shot! We think this tool will transform your linguistic-visual process both in terms of creative power and discovery."
Amal Jos Chacko
Midjourney header
Midjourney header

Midjourney 

San Francisco-based independent Artificial Intelligence research lab Midjourney unveiled their new “/describe” feature that transforms images into words in a tweet. The company, popular for its AI-fueled ability to create images based on a series of prompts, launched more features including “repeat” and “permutations” for its pro subscribers.

Paul DelSignore, creative technologist and artificial intelligence aficionado, took to Medium to break down how these could benefit users. He envisions a future with better search engine indexing and search functionality as a result of “/describe”. 

This could also go a long way in improving accessibility by automating ALT text generation for images uploaded to the web and could prove to be a godsend for people with visual impairments and reading difficulties in finding their way through the internet.

 Midjourney permits the user to upload an image upon encountering the “/describe” prompt and then presents the user with four descriptions its AI engine generates based on the image provided. 

These descriptions are followed by buttons numbered one to four- representing each generated description. These buttons remix the image based on its corresponding description, should the user click on it. 

DelSignore further shares his experience of remixing the official crew portrait of NASA Astronauts for the Artemis II space program.

Midjourney V5 rolls out with new features promising accessibility
Crew portrait of the Artemis II

 

Midjourney described the crew portrait as “NASA astronauts group pose for a photo, in the style of photorealistic portraits, dark cyan, and orange, uniformly staged images, romantic depictions of historical events, celebrity portraits, Hasselblad h6d-400c, non-representational — ar 117:77 — v 5”

The star of the show, however, was the remixed image Midjourney generated based on this description.

Midjourney V5 rolls out with new features promising accessibility
Midjourney generated image based on a description generated from the crew portrait of the Artemis II program

However, not all images get accurate descriptions. While testing the feature with a portrait of a woman in a polka dot dress, PetaPixel observed Midjourney come up with descriptions of varying quality. 

From misidentifying the lens used to capture the picture to mislabelling the dress to be black and white checkered, these descriptions fluctuate from interesting to puzzling, sprinkled with differing levels of accuracy. Images of landscapes resulted in generally useful descriptions but wrong location information.

Although in an early stage, and far from the finished product, Midjourney’s "/describe" tool is intriguing and shows promise, concluded PetaPixel

Equally promising is the new “permutations” feature that allows pro subscribers to create up to 40 combinations in a single query, by listing options within curly brackets {} along with the “/imagine” prompt.

Permutations, along with “repeat”, another new feature that runs the query many times specified with the “--repeat” parameter, provide improvements that, although seemingly small, will be embraced by power users.

Add Interesting Engineering to your Google News feed.
Add Interesting Engineering to your Google News feed.
message circleSHOW COMMENT (1)chevron
Job Board