Midjourney, a leading artificial intelligence image generation platform, has made a surprise update to its core model. In version 6.1, human skin looks more natural and rendered text is more readable.

There was speculation that the company would release v6.5 at some point this summer, but it appears that they have taken a more iterative approach and v6.2 will be released within a few weeks.

The biggest changes relate to the person. Specifically, the way arms, legs, hands, and torso depictions are handled. Texture mapping will also be upgraded, providing new skin textures.

After playing around a bit with the new version, 6.1 feels more like a major upgrade than its repetitive numbering would suggest. The improved text rendering alone is worth the change.

Overall, changes have been made to every aspect of the model. Subtle upgrades in each area have improved image quality by reducing the number of pixel artifacts, enhancing textures, and improving the way certain styles, such as 8-bit and retro designs, are handled.

Midjourney states that the new model is 25% faster when running standard jobs, and the upgrade to the personalization model allows for improved nuance, surprise and accuracy over v6

The company writes on X says: "V6.1 comes with an all-new upscaling and personalization model that significantly improves image quality, consistency, and text. Enjoy the best model yet"

.

One of the new changes is the way upscaling works, providing better image and texture quality to improve the overall look and feel. There is a new -q 2 mode, which takes longer but adds more textures to further improve the realism of the image.

There are also more accurate, detailed, and correct small image features, perfect for eyes, small faces, distant hands, etc.

The feature I am most excited about is improved text accuracy. This is where all AI models suffer, but according to Midjourney, if you put a word in the quote in question, it will render that word accurately on the image.

To use Midjourney v6.1, simply add -v 6.1 to the end of the prompt. This works for the web and Discord versions and changes the model used. We have done some testing and the most obvious change is the skin and text rendering.

The first test gave the prompt: a poster for a movie called "Cats in Space," with the subheading "They're Cats Good" showing a cat on the moon in a space suit. This was detailed enough to dictate the model and included textual requirements.

The poster turned out better than expected, but only two versions had the correct style, one of which had the headline and subheadings correctly drawn.

I next requested that a "wide shot of a woman playing a free piano at a train station" be shown. This was so vague that if the next prompt had been off, it would have shown something odd, but it did not, and one version showed exactly what I wanted, although she was on the railroad tracks.

Finally, I had Midjourney v6.1 generate an image of the woman and animate it using the impressive new Runway Gen-3 Alpha image-to-video feature.

Overall, this is a marked improvement in Midjourney, providing subtle but important changes where the base model was struggling, and is a great sign of what is to come in v7.