On Wednesday, Midjourney announced version 5 of its commercial AI image synthesis service, which can produce photorealistic images at a quality level that some AI art fans are calling creepy and “too perfect.” Midjourney v5 is available now as an alpha test for customers who subscribe to the Midjourney service, which is available through Discord.
“MJ v5 currently feels to me like finally getting glasses after ignoring bad eyesight for a little bit too long,” said Julie Wieland, a graphic designer who often shares her Midjourney creations on Twitter. “Suddenly you see everything in 4k, it feels weirdly overwhelming but also amazing.”
Wieland shared some of her Midjourney v5 generations with Ars Technica (seen below in a gallery and in the main image above), and they certainly show a progression in image detail since Midjourney first arrived in March 2022. Version 3 debuted in August, and version 4 debuted in November. Each iteration added more detail to the generated results, as our experiments show:
Ars Technica
Midjourney works similarly to image synthesizers like Stable Diffusion and DALL-E in that it generates images based on text descriptions called “prompts” using an AI model trained on millions of works of human-made art. Recently, Midjourney was at the heart of a copyright controversy regarding a comic book that used earlier versions of the service.
-
An AI-generated “synthetic photograph” of a girl through a window generated using Midjourney v5 by Julie Wieland.
Julie Wieland -
An AI-generated “synthetic photograph” of a cheeseburger generated using Midjourney v5 by Julie Wieland.
Julie Wieland -
An AI-generated “synthetic photograph” of a boy and flowers generated using Midjourney v5 by Julie Wieland.
Julie Wieland -
An AI-generated “synthetic photograph” of a clown generated using Midjourney v5 by Julie Wieland.
Julie Wieland -
An AI-generated “synthetic photograph” of a girl generated using Midjourney v5 by Julie Wieland.
Julie Wieland -
An upscaled version of a Midjourney v5 output with the prompt “a muscular barbarian with weapons beside a CRT television set, cinematic, 8K, studio lighting.”
After experimenting with v5 for a day, Wieland noted improvements that include “incredibly realistic” skin textures and facial features; more realistic or cinematic lighting; better reflections, glares, and shadows; more expressive angles or overviews of a scene, and “eyes that are almost perfect and not wonky anymore.”
And, of course, the hands.
Just a heads-up – Midjourney’s AI can now do hands correctly. Be extra critical of any political imagery (especially photography) you see online that is trying to incite a reaction. pic.twitter.com/ebEagrQAQq
— Del Walker (@TheCartelDel) March 16, 2023
Over the past year, the idea that AI art generators can’t render hands correctly has become something of a cultural trope. Notably, Midjourney v5 can generate realistic human hands fairly well. “Hands are correct most of the time, with 5 fingers instead of 7-10 on one hand,” said Wieland.
In the service’s Discord release notes, Midjourney also noted that v5 now responds with a “much wider stylistic range” than version 4, while also being more sensitive to prompting, generating less unwanted text, and offering a 2x increase in image resolution.
If there’s a visual downside to the Midjourney upgrade for AI art fans, it perhaps comes from images that can be so realistic and “perfect” that the model’s precision takes away some of the thrill of repeatedly generating AI imagery to find a suitable result, what one might call a “slot machine effect.” Although one Twitter user named Philipp Lenssen noted, “If you have a specific image subject in mind, it’s still a bit like lottery. But with higher winning chances than v4.”
