Midjourney, the leading artificial intelligence image generation platform, has surprised its users with the release of an update to its core model. Version 6.1 brings significant improvements to the platform, particularly in the rendering of human skin and text legibility. While there were speculations about the release of v6.5 this summer, the company has opted for an iterative approach, with v6.2 expected in the coming weeks.
The most notable changes in Midjourney v6.1 are focused on the depiction of arms, legs, hands, and bodies, resulting in more natural-looking representations. Additionally, the texture mapping has been upgraded to offer new skin textures, further enhancing the realism of the generated images. Users who have had the opportunity to try out the new version have expressed that it feels like a major upgrade, surpassing the expectations set by its iterative numbering. The improved text rendering alone is considered a significant improvement.
Midjourney v6.1 introduces changes across every aspect of the model, with subtle upgrades aimed at enhancing image quality by reducing pixel artifacts and improving texture handling, including specific styles such as 8-bit and retro designs. The company claims that the new model runs 25% faster for standard jobs, and the personalization model has been improved to provide better nuance, surprise, and accuracy compared to v6.
The new upscale feature in Midjourney v6.1 offers better image and texture quality, resulting in an overall improved look and feel. The addition of a new “-q 2” mode allows for a longer processing time but adds more texture, further enhancing the realism of the generated images. Moreover, the model now accurately renders text within quotations, addressing a common challenge faced by AI models.
Initial tests with Midjourney v6.1 have shown promising results. Users have reported significant improvements in skin and text rendering. For instance, when prompted to create a poster for a movie titled “Cats in Space” with the subheadline “They are feline good,” the generated poster exceeded expectations, with accurate rendering of the headline and subheadline. Similarly, when asked to generate a wide shot of a woman playing a public piano in a train station, the model delivered the desired result, showcasing its ability to understand vague prompts.
Furthermore, when combined with the impressive new Runway Gen-3 Alpha image-to-video functionality, Midjourney v6.1 produced one of the most realistic AI-generated images and videos to date. This notable improvement in Midjourney demonstrates the platform’s commitment to addressing areas where the base model previously struggled.