AI imager Midjourney v5 stuns with photorealistic images—and 5-fingered hands

wednesday on the way announced Version 5 of a commercial AI image synthesis service capable of producing photorealistic images at the level of quality demanded by some AI art fans Spooky and “too perfectMidjourney v5 is currently available as an alpha test for customers who subscribe to the Midjourney service available through Discord.

Graphic designer Julie Wieland said, “After ignoring poor eyesight for a little too long, the MJ v5 feels like it’s finally got glasses.” often share Her mid-journey work on Twitter. “Suddenly everything is in her 4k. It’s strangely overwhelming, but also amazing.”

Wieland has shared some of the Midjourney v5 generation with Ars Technica (see gallery below and main image above). Since Midjourney first appeared in his March 2022 appearance, the details in the image have definitely progressed. Version 3 debuted in his August and version 4 debuted in November. As our experiments show, more detail was added to the results generated for each iteration.

Midjourney works similarly to image synthesizers such as Stable Diffusion and DALL-E, using AI models trained on millions of man-made works of art to generate images based on textual descriptions called ‘prompts’. To do. Recently, Midjourney was at the center of a copyright dispute over a comic book that used previous versions of the service.

After trying v5 for a day, Wieland noticed improvements, including “incredibly realistic” skin textures and facial features. More realistic or cinematic lighting. Better reflections, glare, and shadows. A more expressive angle or overview of the scene, and “near-perfect, no-wobbly-eyes”.

Hands, of course.

Over the past year, the idea that AI art generators can’t render hands correctly has become something of a cultural metaphor. In particular, Midjourney v5 does a pretty good job of generating realistic human hands. “Hands are mostly one-handed and he’s right with 5 fingers instead of 7-10,” Wieland says.

In the service’s Discord release notes, Midjourney notes that v5 is now responsive with a “much wider stylistic range” than version 4, being more responsive to prompts, generating less unnecessary text, and reducing images You also mentioned that the resolution has been doubled.

For AI art fans, if there’s a visual downside to the Midjourney upgrade, it’s probably due to the extremely realistic and “perfect” images, where the model’s accuracy requires AI images to find the right results. It can take away some of the thrills you generate over and over again. Some might call it the “slot machine effect”. There is a Twitter user named Philipp Lenssen, I got it“If you have a particular image subject in mind, it’s a bit like the lottery, but with better odds of winning than v4.”