Tech

Midjourney version 5 improves almost every aspect of AI-generated images.

2 weeks ago
0

To become better: With all the latest news revolving around ChatGPT and other big language models, it’s easy to forget that their cousins, AI image generators, are still getting better. Perhaps someone has figured out how to render eyes and hands without making the object look like something out of a nightmare. however, the results still scare some people.

Earlier this week research lab middle of the road has released a beta version of the fifth version of its eponymous AI imaging software. According to the announcement on Twitter, the latest version adds higher image quality, more “diverse” results, a wider range of styles, seamless textures, and more.

Users have already posted hundreds of stunning results, and opinions about the improvements are mixed. Most of them are impressed that the artificial intelligence has struggled to reproduce aspects such as shadows, reflections, eyes and hands. Below is an image we created with OpenAI Dall-E as an example of where the machine is having issues.

Read More

The composition is somewhat odd and the overall feel is cartoonish. Not so with lighting. The eyes and hands are severely deformed. The legs are filthy with artifacts, as are the popcorn container and the seat next to the subject. This result one of four with similar problems to varying degrees.

Version 5 of Midjourney seems to have improved in this regard, at least from the examples others have shared. The results of simple clues border on the uncanny valley – realistic enough to pass for professional photographs in many cases, but still with that odd quality that you can’t pinpoint. Even though the images are incredibly realistic, many have called them creepy.

Our very own Kishalay Kundu said, “I’m honestly more scared than impressed” after viewing a series of nearly flawless photos of the Midjourney V5. The fear was that it was quite easy to create a fake image and pass it off as real.

Besides the creep factor, compared to the V4, the Midjourney V5 has a much improved quality. Graphic Designer Julie Wieland used Midjourney V4 for a while (released last November) and says version 5 has “incredibly realisticSkin textures. Lighting effects are also much better, including reflections, highlights and shadows. Perhaps most importantly, the AI ​​generates hands and eyes that look natural most of the time.

“The eyes are almost perfect and no longer wobble,” Wieland told Ars Technica. “Hands are mostly correct, with five fingers instead of 7-10 on one hand. MJ v5 currently feels like it finally got points after ignoring poor eyesight for too long. All of a sudden you see everything in 4k; it seems strangely overwhelming, but also surprising.”

Midjourney also improved the native resolution from 512x512px to 1024x1024px. The increase aligns it with Dall-E. However, version 4 can supersample to double the original resolution. It’s perfectly reasonable to expect V5 to use the same technique to create 2048×2048 images, but that will be an update in the future.

The bottom line is that MidJourney entered the AI ​​scene just a year ago. Many (not all) of these images that have flooded Twitter this week have remained untouched. Weiland previously used a combination of techniques to improve the visual quality of Midjourney 4, including “painting” with Dall-E and retouching in Photoshop. Version 5 promises less post-generation editing and possibly perfect photos faster than we can imagine. This prospect is truly both exciting and terrifying at the same time.




Source link

2 weeks ago
0

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button