Dance of the Machine: Midjourney’s AI creations already look like the real thing

Photorealism records are broken in the visuals of Midjourney 5.0, to the point where it is not always clear what is real and what is artificial. Natural language understanding has also been significantly improved. What will happen to art now?   

Even in a world saturated with Generative AI – those tools that are capable of generating texts and images on their own based on general instructions – the Midjourney software in its latest version (5.0) is jaw-dropping.

It starts with the image quality, with new records of photorealism: the new renderings that have begun to flood the networks look in part like authentic photographs for all intents and purposes, with the boundaries between reality and computer processing blurred (even the strange seven-finger problem has been solved, and today the software already understands that most of us only have five).

The new version provides more diverse products and supports wider dynamic ranges and variable image sizes. It understands natural language sentences better, so there is no longer a need for multiple descriptions and distinct words separated by commas or slashes (“Draw me a unicorn floating in the night sky”, rather than: “unicorn, floating, night”).

Founded by David Holtz, Midjourney was born as a competitor to popular image creation models such as OpenAI’s DALL-E. To use Holtz’s model, the user enters a text input (Prompt) that details their request and immediately receives four images – four visual representations of the prompt they gave. From here, they can further improve, correct, and style it.

Unlike its main competitor, Midjourney opened its model to the general public from the very beginning. However, Holtz’s tool is not accessible as a web or mobile application, but only through an independent server developed by the company (a Discord server). In this way, it hopes to achieve a collaborative experience driven by the power of the community, when the results can be changed jointly with the other users on the server – and not just be excited about the miracle that others have done. After 25 free uses, each user is required to purchase a monthly subscription at a cost of $6-80, depending on the chosen route.

The criticism of Midjourney concerns the fact that, unlike OpenAI’s tools, it has almost no built-in barriers. Thus, it can be used to produce visuals that mock celebrities, rely on stereotypes, or quite convincingly recreate the style of artists, by training the algorithm on an archive of their works. A class action lawsuit was recently filed against the company and others like it on the subject. According to reports, Midjourney’s algorithm was trained using works without permission, which immediately raised concerns about copyright. On the other hand, nude images, which were previously compared to porn, are no longer possible in later versions.

The ability for any user to create works of art and realism from scratch is a new land of limitless possibilities. In the world of digital advertising and marketing, tools like Midjourney open up a wide range of opportunities, from shortcuts for designers to a welcome refresh for illustrations. But that’s a topic for another article.