New Midjourney V5 Generative AI Offers Photorealistic Images and Hands With 5 Fingers
gpt-4 is cool but midjourney v5 can now do hands pic.twitter.com/VwkCIZV7yh
— fabians.eth (@fabianstelzer) March 16, 2023
Midjourney has released version 5 of its synthetic image generator, offering photorealistic images that are far less likely to include jarringly wrong details like hands with too many fingers or bizarre joints. The alpha test for V5 is available to those who sign up and pay for a Midjourney subscription on Discord.
Midjourney originally debuted in March last year and quickly iterated to version 3 by August, when it became more well-known and widely used. Version 4 came out in November and a few months later, we are version 5. Each new edition of Midjourney improved details, resolution, accuracy, and overall quality. At the moment, V5 actually generatess images more slowly than V4 on the usual four-image grid system, despite no access to ‘relax mode,’ just ‘fast mode.’ That’s because V5 images are produced already upscaled, with enhanced details visible from the get-go, rather than requiring users pick individually from the grid which images to improve. Hands and other limbs are very noticeablly improvement, though only anecdotally. As seen in the top image, the new Midjourney can reliably make mostly realistic hands, a big step up for the generative model. Even besides the question of hands, Midjourney’s new model could easily pass off some of its scenes and characters for real life, a far cry from what it used to offer. You can see a comparison of, from left to right, Midjourney V3, V4, and V5.
street photo in new york, medium shot, natural lighting, shot on fujifilm, detailed and realistic environment, cinematic –ar 2:1
— phil desforges ⟠ (@storybyphil) March 15, 2023