The AI image generator Midjourney introduces a new code base and AI architecture with V4. The generated images are more detailed, concrete and can be more realistic.
AI image generator Midjourney has released a new alpha version of its algorithm. David Holz, CEO of Midjourney, announced several changes that are introduced with V4.
The new version features
- significantly more knowledge, for example, about creatures or places,
- is better able to recognize small details correctly,
- and can process more complex prompts.
- V4 also handles scenes with multiple objects and characters better.
According to Holz, V4 has an entirely new code base and AI architecture. It is the first model trained on the Midjourney AI supercluster, which was developed in the past nine months.
Holz describes V4 as a first step, not the last, and the beginning of something profound and unfathomable. Wood had previously predicted a bright future for generative AI, all the way to fully AI-generated video games within a decade.
Midjourney enforces content guidelines more strictly because images are now more realistic
Midjourney V4 can create much more realistic images, according to David Holz, but that also comes with risks. From now on, community guidelines would therefore be enforced much more strictly. So anyone who obviously tries to bypass the list of forbidden words in a prompt risks being banned sooner rather than later.
In general, V4 would also require an entirely new way of formulating prompts. The remix function, i.e., the possibility to further edit generated images via AI, is already enabled for v4.
While V4 is a step forward and provides much more appealing results according to some users, there are also still limitations in the alpha version. Probably the biggest is the limitation to a square aspect ratio.
Also on the developers' to-do list are higher resolution and image quality, upscaling control, and image sharpness. On the other hand, problems with text artifacts and cropping errors should be eliminated.
The quality raises, but woth quality also kitsch factor increases. It's all in the prompts, which should be written differently, as devs state. pic.twitter.com/egOP6nkisc
— Merzmensch Kosmopol (@Merzmensch) November 5, 2022
Additional improvements for Midjourney V4
Shortly after the release of V4, Midjourney announced a few more upgrades, mostly to the web interface. Fast forward:
- In the web interface, images can now be sorted into collections.
- Collections and highly rated images are displayed at the top of the profile.
- Profiles can be customized by username, profile and header images.
- Widescreen mode, tile preview, rating filters and custom home filters added to web interface.
- Batch publish/remove and add to collections added in web interface.
- Batch download size can be customized in the web interface.