Midjourney version 5 improves nearly every aspect of AI-generated images

improve: With all the recent news swirling around ChatGPT and other large paradigms of languages, it’s easy to forget that their cousins ​​- AI image generators – are still getting better. One might have figured out how to make eyes and hands without making the subject look like something out of a nightmare. However, the results still creep up on some people.

Earlier this week, the research lab Medjourney Released a beta for version 5 of the AI ​​imaging software. According to its announcement via Twitter, the latest version adds higher image quality, “more varied” results, a wider range of styles, smooth textures, and more.

Users have already posted hundreds of amazing results, and feelings about the improvements have been mixed. Most of them were impressed because AI photography struggled to reproduce aspects like shadows, reflections, eyes, and hands. Below is an image we created using OpenAI’s Dall-E as an example of where the device is having a problem.

The combination is somewhat off, and the overall feel is cartoonish. The lighting is all wrong. The eyes and hands are severely deformed. The legs are contaminated with artifacts, as are the popcorn container and seat next to the subject. this result one of four With similar problems to varying degrees.

Midjourney version 5 seems to have improved in this regard, at least from the examples others have shared. The results from simple claims border on the uncanny – realistic enough to pass off as professional photos in many cases, but still with such an eerie quality that you can’t quite put it down. Despite being incredibly realistic, many have described the images as creepy.

“I’m more scared than impressed, to be honest,” said our very own Keshalaya Kondo after viewing a series of photos of the nearly flawless Midjourney V5. The fear is that one can easily create a fake photo and pass it off as genuine.

Creep factor aside, compared to the V4, the Midjourney V5 has greatly improved quality. Graphic designer Julie Weiland Used Midjourney V4 (released last November) for some time and it says v5″Incredibly realisticskin texture. The lighting effects are also much better, including reflections, glow, and shadows. Perhaps most importantly, the AI ​​generates hands and eyes that look normal most of the time.

“The eyes are almost perfect and no longer wonky,” Weiland told Ars Technica. Hands are correct most of the time, with 5 fingers instead of 7-10 on one hand. Currently the MJ v5 feels like I’m finally wearing glasses after ignoring bad eyesight for far too long. Suddenly you see everything in 4K; it feels weirdly disorienting but amazing also “.

Midjourney has also improved the original resolution from 512 x 512 pixels to 1024 x 1024 pixels. The increase is in line with Dall-E. However, v4 can replace the sample to double the original resolution. It’s not unreasonable to expect the V5 to use the same technology to produce 2048 x 2048 resolution images, but that’s for another update in the future.

The bottom line is that MidJourney only hit the AI ​​scene a year ago. Many (but not all) of these photos flooding our Twitter feeds this week are untouched. Previously, Weiland used a range of techniques to improve Midjourney 4’s visual quality, including “outside coloring” using Dall-E and touch-ups in Photoshop. Version 5 promises less post-generation editing and possibly picture-perfect images sooner than we can imagine. This prospect is both exciting and frightening indeed.

Source link

Related Posts