Midjourney has announced that its latest version, V5, will be released this week.
Users are thrilled because they know what’s coming: last week, Midjourney invited its community to a rating party to collect data about the newest model (the party has since ended).
While Midjourney CEO David Holz says these preview images are “not representative of the V5 algorithms”, we can guess (just like with the V4 rating party) that they’re pretty damn close.
So what’s new?
Apart from a big increase in resolution, what stands out most is a big increase in the quantity and quality of practical use cases.
Midjourney has a treasure trove of data on what it’s users are trying to make. We’re guessing that real world use cases were a huge underserved need in the last version.
This underscores a bigger transition in the audience for AI products this past year, from hobbyists to professionals.
So what do we have to look forward to this week?
Here’s a roundup of what impressed us the most:
Midjourney strikes back just when people were talking about how custom Stable Diffusion models had gotten ahead of Midjourney in the realism game. Just look at those hands!
One of the coolest ways people used Midjourney was to publish children’s books. What surprised us was how clean the illustrations have become in V5, and how much closer they are to being used directly, without any post-processing.
Game Assets & Concept Art
|Wow that’s a detailed rock.|
A large number of game studios are already using Midjourney in their workflows. The range and quality of assets they will be able to work with will increase greatly.
V5’s range now spans from antiquity to the present – everything from Greek Frescoes to Picasso imitations.
We’re getting closer and closer to things designers and developers will actually be able to use. Fidelity, clean-ness and composition are all greatly improved!
These illustrations left us breathless. We’re at this point now: these piece are indistinguishable from the art used to create AAA games.
Did you know that employees of the prestigious Zaha Hadid Architects use Midjourney and prompt “in the style of Zaha Hadid” in their work? In the new model we’ll see capabilities for more specific types of architectural deliverables.
The details, depth and lighting of photorealistic environments is astonishing. Cue people making fake images to put on Google Maps?
Realism as a whole takes a big step forward. This will be the model that creates results that people won’t be able to distinguish from reality.
‘Weird’ and abstract art styles, which were underrepresented in the previous version, are big in this one! We’re looking forward to all of the crazy experiments that people will try.
Keep your eyes on the Midjourney Discord this week.
If the V5 release is anything like the last one, it will drop at a random and strange time (V4 was released 1AM EST…thanks David Holz).
We’ll post more updates on Twitter as they come in.
Till next time,
Yubin & Crew