Midjourney, one of the first AI-based image generators, has received a new version.
The model Midjourney V7 is currently in alpha and despite the fact that it is not optimized for viral images in the style of Studio GhibliHer attempts look at least aesthetically pleasing.
Before you get started, you’ll be asked to evaluate about 200 images to «personalize» your profile (V7 is the first Midjourney model to have personalization enabled as standard) and customize the model to work with your visual preferences.
One of the most exciting new features for our new V7 model is something we call “Draft Mode”. Draft mode is half the cost and 10 times the speed and it might be the best way to iterate on ideas ever. Try it with voice, think out loud and let our ideas flow like liquid dreams. pic.twitter.com/ANfTMC6Ej1
— Midjourney (@midjourney) April 4, 2025
Other major updates include the draft mode, which costs half the price and speeds up the process by 10 times. If you like the draft, you can regenerate it in normal quality.
According to Midjourney CEO David Holtz, V7 has become smarter in working with text hints, while bodies and previously «problematic» limbs now look more detailed.
nice hands
— Midjourney (@midjourney) April 4, 2025
At the same time, some standard neural network features are still not available for V7, including image scaling — they are promised to be added within the next two months.
«This is a completely new model with unique advantages and probably a few disadvantages,» Holz wrote on Discord. «We want to hear from you about what makes it good and what makes it bad, but keep in mind that different styles of prompts are worth experimenting with. So play around a bit».
Midjourney — is unusual in that when it was founded by Holtz in 2022, it did not raise a single penny of outside money. At the end of 2023, the San Francisco-based company expected about $200 million in revenue, but a lot has changed since then, including the number of competitors. Holtz recently stated that they are building a hardware team to work on some projects, but without details, and are focusing on training the previously announced models to create videos and 3D objects.
