AI art generator Midjourney began alpha testing version 4, what is the result? Explore here

Midjourney v4 is the new version of an AI algorithm designed to create award-winning, weird, and uncanny graphic compositions. The developers are alpha testing their latest tech while working on a proper full launch of the final product. Machine learning algorithms designed to generate uncanny or bizarrely surrealistic images are taking the tech world by storm, and things will only get more complicated in the future. Case in point: there's a new version of Midjourney –a "totally new" AI architecture –which is just the beginning of a journey towards "something deep and unfathomable."

On Saturday, AI art generator Midjourney began alpha testing version 4 ("v4") of its text-to-image synthesis model, which is available for subscribers on its Discord server. The new model provides more detail than previously available on the service, inspiring some AI artists to remark that v4 almost makes it "too easy" to get high-quality results from simple prompts.

Midjourney opened to the public in March as part of an early wave of AI image synthesis models. It quickly gained a large following due to its distinct style and for being publicly available before DALL-E and Stable Diffusion. Before long, Midjourney-crafted artwork made the news by winning art contests, providing material for potentially historic copyright registrations, and showing up on stock illustration websites (later getting banned).

Over time, Midjourney refined its model with more training, new features, and greater detail. The current default model, known as "v3," debuted in August. Now, Midjourney v4 is being tested by thousands of members of the service's Discord server who create images through the Midjourney bot. Users can currently try v4 by appending "–v 4" to their prompts.

Like with DALL-E, Stable Diffusion, and other AI algorithms, Midjourney can take a simple or very complex text prompt and turn it into a composite image. Led by David Holz, who previously founded the Leap Motion motion control venture, the Midjourney team was recently able to stand out in the crowd of image-generating algorithms by winning the Colorado State Fair's annual art exhibition. Midjourney v4 can generate much more realistic pictures than previous models, so much so that the team was forced to adapt stricter rules about images that can "cause drama." The new Midjourney algorithm was trained on a new AI supercluster and has been in the works for over nine months. Users should expect quirks and inconsistencies during the alpha test while the team is pushing significant changes toward full launch.

In the tests of Midjourney's v4 model, it is founded that it provides a far more significant amount of detail than v3, a better understanding of prompts, better scene compositions, and sometimes better proportionality in its subjects. Some results can be difficult to distinguish from actual photos at lower resolutions when seeking photorealistic images.

However, Midjourney says that v4 is still in alpha, so it will continue to fix the new model's quirks over time. The company plans to increase the resolution and quality of v4's upscaled images, adding custom aspect ratios (like v3), increasing image sharpness, and reducing text artifacts. Midjourney is available for a monthly subscription fee that ranges between US$10 and US$50 a month.