Midjourney V5 Is Finally Here! How Does It Stack Up?
Better late than never: Midjourney version 5 is out at last! I take it for a spin and run a comparison between V5 and the previous versions of Midjourney.
Two months later than expected, but Midjourney V5 has officially landed.
The raw, alpha version 5 was released to everyone on Discord yesterday, which means you—yes, you—can test it out for yourself right now!
But what’s changed in this version and how does it compare to what came before?
Let’s go right ahead and find out!
Important note: The current alpha release is still being tested and fine-tuned. The final, stable release of V5 may end up having a different default aesthetic. I’ll make sure to update this post with any future changes.
Edit 04-05-2023: The new V5.1 is here!
What is Midjourney V5?
Midjourney version 5 is the latest iteration of Midjourney’s text-to-image algorithm, which is regularly updated and improved. The chronology so far has looked like this:
Version 1: March 2022
Version 2: April 2022
Version 3: July 2022
Version 4: November 2022
Version 5: March 2023
As you can see, the first few versions were relatively quick iterations with a longer gap before Midjourney V4.
Version 4 was a massive leap forward in terms of realism, image cohesion, and general aesthetic. After that, Version 5 was yet another long wait.
So…what’s changed?
What’s new in Midjourney V5?
In short, a whole lot.
You’ll need a Discord account to read the official FAQ I linked to, but don’t worry: I’ll cover all of it right here in more detail.
1. Higher native resolution
Midjourney Version 5 images are by default rendered at twice the resolution of V4—1024 X 1024 px compared to 512 X 512 px. V4 images had to be upscaled to reach that resolution, so this is a huge improvement.
Later, we might see V5 upscalers that can further increase the resolution to e.g. 2048 X 2048 px.
But for now, 1024 X 1024 px is both the default and max output.
2. Unlimited aspect ratios
For comparison, Midjourney V4 only allowed the square aspect ratio (1:1) at launch, which later increased to 2:3 / 3:2 until finally supporting resolutions of up to 1:2 / 2:1.
3. Sharper, more detailed images
Images produced by V5 are notably sharper and have more detail.
Here’s a comparison of V4 vs. V5 for the prompt: “photo of two toddlers playing with LEGO”
At first glance, the V4 image looks real enough. But that quickly falls apart upon closer inspection (click to enlarge the images). Children’s faces look more like rubber masks, their hands are a mess, and the LEGO pieces are really just vague replicas.
V5 straight up looks like a real photo. The level of detail is truly striking.
4. More coherent output
On a somewhat related note, Midjourney V5 has also improved the overall composition of images and fixed a number of pesky issues in V4, including:
Better handling of large groups of people
More realistic hands with the right amount of fingers (most of the time)
Fewer random artifacts in images
Here’s a “candid photo of a crowd of people at a backyard barbecue”:
Again, the images appear similar on the surface, but a careful look reveals weird hands, warped faces, and other “off” details in V4.
Version 5—while not flawless—has fewer oddities and is better at rendering distinct faces.
5. Wider range of supported styles
The team behind Midjourney claim V5 has five times the stylistic range of V4.
In theory, this means that you can coax all sorts of styles and specific visuals out of version 5 with proper prompting. Version 5 is apparently particularly great at rendering landscapes and architecture. Here’s a quick test:
6. More nuanced understanding of text prompts
MJ V5 should also be better at natural language processing (NLP).
This means you can try longer and more descriptive prompts written in complete sentences to get what you want. (Not to be confused with splatterprompting.)
7. Support for tiling
Tiling is back, y’all!
This was a feature of V3, but V4 never supported it. This option generates “tiles” that can create an interlocking, recurring pattern.
Take an emoji prompt like 🌈. Without the tile modifier, you may get a regular photo image like this:
By adding the “--tile” parameter to your prompt, you might get this:
You can now string many of these “tiles” together into a seamless chain. Neat!
8. Better handling of image prompts (and remixes)
Using image prompts with Midjourney Version 5 will generally result in more “accurate” and predictable outcomes.
Here’s an example. I took the following Unsplash image of a tree:
Then I used it as an image prompt in Midjourney in combination with a text prompt saying “Pixar cartoon.” Let’s compare how V4 and V5 handled this.
Here’s the 4-image grid from Version 4:
And here’s the 4-image grid from Version 5:
As you can see, Version 5 sticks more faithfully to the composition of the original image and stays largely consistent in its rendering of the tree.
This is supposed to carry over to the “Remix” feature, which fuses two (or more) input images into one.
To test this, I tried blending the following two Unsplash images:
I threw them into V4 and V5, without any additional text prompts.
Here’s the V4 grid:
Here’s the V5 one:
Honestly? The jury is out on this one.
V4 definitely pulls the resulting image into the “toy” direction while V5 is more realistic (and better at keeping the man centered.)
Which one you prefer depends on your initial intent. Perhaps with text prompts included, it’d be easier to assess whether V5 is indeed more accurate.
9. Support for image weights
This is yet another feature from V3 that wasn’t available in V4.
Well, it’s returned in V5!
It lets you decide how much priority Midjourney gives to the image compared to the text part of the prompt.
You assign image weights by evoking the “--iw” parameter, followed by a number between 0.5 (lowest weight) to 2 (highest weight). Like so:
To show the impact this has, I used the following as the image prompt:
The text part of my prompt was “happy face.”
Here’s how Midjourney V5 combines them when I use --iw 0.5 (lowest):
As you can see, V5 takes only vague inspiration from the sun’s position and returns a variety of rather different “happy faces.”
Now watch what happens with --iw 2 (highest):
The sun prompt now clearly takes center stage, even at the expense of rendering any faces whatsoever in two of the images.
Image weights give you a powerful way to experiment and find just the right balance between text and image input. Give them a try!
10. No built-in aesthetic (yet)
Midjourney crew refers to this as V5 alpha being “less opinionated.”
What does that mean?
Anyone who used V4 for a while knows that it defaults to a certain distinct “look” in the absence of additional descriptors. This often ends up being a close-up of a woman’s face that is almost-but-not-quite realistic (a bit like a computer game character).
Let me show you. Take a one-word prompt like “joy” and throw it into V4:
Clearly, Midjourney V4 adds some secret sauce to our simple prompt behind the scenes to keep this level of consistency across images.
Now take that same “joy” prompt and throw it into V5:
Still plenty of women, but we’re getting a whole range of styles and image compositions. If you want the woman’s face from V4, you’ll have to prompt it more deliberately (or simply switch back to V4 if that’s your thing.)
This is a bit of a double-edged sword.
On the one hand, V5 gives you free rein to craft your exact vision using detailed and specific prompts. On the other hand, you can no longer rely on Midjourney to spit out predictable outputs with little effort on your part.
To use a shitty analogy: V4 is like Mr. Potato Head and V5 is like Play Doh.
You can certainly rearrange Mr. Potato Head’s face in many ways, but he’ll always be Mr. Potato Head.
With Play Doh, you can theoretically create any face (or other object), but it does require more skill to get the result you want.
Note: Midjourney made it clear that the upcoming “stable” version of V5 will also have its own built-in aesthetic. (The current “raw” version will remain available.) But for now, you might have to learn a few V5 whispering tricks.
Edit 04-05-2023: The new, more artsy V5.1 is here!
How does V5 output compare to the previous versions?
Here’s a look at how the different Midourney versions handle the following simple prompts: “forest hut,” “hamster photo,” “hoverboard.”
As you can see, Midjourney Version 4 was already a major leap forward in image cohesion compared to what came before.
Version 5 is a more subtle improvement that isn’t quite as dramatic. But the level of detail and consistency is generally higher, and V5 tends to skew towards photographic images without other descriptors to guide it.
How to get started with Midjourney V5?
If you’ve never used Midjourney before, check out my step-by-step “getting started” guide (mostly steps 1-3).
If you’re already a Midjourney user, you have two options:
Option 1 (“permanent”):
This one’s best if you’re planning to mainly use Midjourney V5 going forward.
Type “/settings” in the chat to open the “Settings” menu:
Click on “MJ version 5” and make sure it turns green.
That’s it! Any image you generate from now on will automatically use the latest version 5.
Option 2: (“per image”)
The alternative is to manually append the “--v 5” suffix to your image prompt:
Stick to this option if you prefer to work with Midjourney V4 (current default) and only want V5 occasionally (e.g. for specific types of images).
Now go have an absolute blast out there!
Over to you…
Have you tried Midjourney V5? How do you think it fares compared to V4? Are there any situations where you’d still prefer older Midjourney versions instead of V4 or V5?
If you want to share your thoughts or cool MJ V5 images, I’m always happy to read (and see) them. Send me an email or leave a comment below.
“Why Try AI” is 100% free to read for everyone. There are no content paywalls or other pay-to-access features.
But you can help support my work through an optional paid subscription. It’s like buying me a fancy cup of coffee once a month. If you do, thank you!
Great post! These kinds of side-by-side comparisons are just what I wanted to see. I don't think anything demonstrates Midjourney's progress quite like that hamster photo. Mutant mogwai to adorable critter in just 1 year!
Tried v5 today with some old prompts. The results (esp. with prompts that had meh outcomes before) are mind blowing. However, I am not a huge fan of the fact that images turn out more photorealistic.