news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Mingmin from Aofei Temple
Quantum Bit | Public Account QbitAI

Midjourney v6.1 was released and became an instant hit!

Various portrait blockbusters, completed in one go.

The pictures, lighting, characters' expressions and movements... are all closer to reality.

The delicate texture of the statue can also be restored in detail.

Compared with the previous version v6, the improvements in all aspects are very significant.

Netizens lamented: This is almost indistinguishable from a photo blockbuster.

Combined with the recent runway Gen-3, someone created a stylish blockbuster with full detail and realism in just 45 minutes.

After half a year, Midjourney brings a major update and is now open for experience.

v6.1 has been upgraded in 8 aspects. In a word, it makes the generated images look better.

And the next version will be released soon. The official said that v6.2 may be available next month and will have more upgrades, such as text.

Now they need to collect a lot of user usage data first. Now when you open Midjourney, the default version is v6.1.

Unrivaled in portraiture

According to the official introduction, this version upgrade mainly has the following improvements:

• Greater consistency (such as arms, legs, bodies, plants and animals, etc.)
• Higher image quality (reduced artifacts, enhanced textures, etc.)
• More accurate, detailed, and precise understanding of small image features (such as eyes, small faces, etc.)
• Better graphics/texture quality
• Faster generation speed, 25% improvement in standard images
• Improve text accuracy
• A new personalization model
• Personalized code version control: you can use the previous version of the code in the new version
• A new -q2 mode that adds more textures, but takes longer (25% longer) and is less consistent

The most prominent change is in the portrait aspect, which is almost impeccable.

Combined with the actual tests of netizens, we can see the above improvements more intuitively.

Relatively speaking, version v6.1 is more authentic.

The generated objects also look more reasonable.

Some prompt words that were previously unavailable in v6 can now be handled with better results in v6.1.

It can fix some of the previously obviously distorted generated elements.

Some surreal synthesized images seem to be more comfortable to look at.

However, there seem to be challenges in generating many people and group portraits.

Someone tested "a group of people doing yoga in a square", and the "people" in the output results were very abstract.

And the errors in generating each person are similar: the wrong number of limbs, extending in the same direction.

In the group portrait generation, the faces of the people in the first two or three rows are relatively normal, but those in the back become distorted and terrifying.

Some people think this is because AI can only draw one person, but does not understand that a group of people is made up of individuals.

It's like asking MJ to draw a tree. It knows how to draw a leaf well, but it doesn't know that the crown of the tree is made up of a bunch of leaves, so it can only draw a few of the most prominent leaves, and the rest will be a mess.

If it is the back of the character, the effect will be slightly better.

Have you tried it? How does it feel?

Reference Links:
[1]https://x.com/midjourney/status/1818342845885141185
[2]https://www.reddit.com/r/singularity/comments/1eg2sjt/midjourney_v61_just_released_and_is_practically/
[3]https://twitter.com/WorldEverett/status/1818378770459619653