Midjourney V6.1 is Here and It’s Impressive
New Upscalers, Improved Image Quality, Faster Generation Times, and More
Midjourney AI is undoubtedly the leading image-generation model, and it never ceases to amaze me with its rapid progress over just a few years. Now, with its latest update, we are seeing images with even greater quality and detail.
To illustrate this, here’s a collage showcasing the advancements across Midjourney versions:
But now, let’s talk more in-depth about V6.1.
What’s new
Improved image quality (enhanced textures, reduced pixel artifacts, etc.)
Improved text accuracy
Faster generation times (about 25% improvement)
New 2x upscalers with better quality
New
--q 2parameter that adds more texture but takes longer to generateNew and improved personalization model
Good to know
You can still use Style Reference codes from V6
You can still use Personalization codes from V6
Inpainting & Outpainting is currently not yet updated to V6.1 (you can use these features but the generations will default to V6)
V6.2 update with more improvements expected soon
A larger V7 update is expected later this year
Improved image quality
With Midjourney AI V6.1, you’ll notice finer details and better texture quality in your generated images. The model’s ability to understand prompts has also improved, making incorporating specific details into your creations easier.
The newest version is also better at generating hands, body parts, and animal anatomy, demonstrating a significantly better overall comprehension.
Let’s look at some side-by-side comparisons with the previous version:
While version 6 was already pretty great, you can see that the image generated with V6.1 has more defined and detailed textures.
Here’s another example:
The model also tends to favor creating beautiful images, making it harder to generate something objectively unattractive. Even when attempting to create less visually appealing subjects, the model will do its best to ensure the final result remains aesthetically pleasing.
While it was perfectly possible to generate realistic hands with V6, the new version seems to be better at the job. And the improved skin textures make the hands look way more realistic.
During my testing, I did encounter the occasional hand with six fingers, but overall, the model has shown noticeable improvement. Generations of hands now appear more realistic and refined.
Improved text accuracy
Being able to generate text directly using Midjourney unlocks a range of use cases that previously required external tools. This improved text accuracy simplifies workflows and expands creative possibilities, allowing for seamless integration of text and images in your projects.
When generating images with text, you should always put the text in quotation marks. Here’s an example prompt:
A note written with a pen on a folded paper napkin reading “Call Me — HungryMinded” — ar 4:3 — v 6.1
Let’s see if V6.1 can handle more complex tasks.
Prompt:
A minimalistic poster reading “Learn to use Midjourney AI today, your friendly tutor is here to help” — ar 4:5 — v 6.1
The result is not perfect, but it’s getting there. And the cool part is that you can always edit the text using the “repaint” feature.
Improved upscaling
The default size of images generated with V6.1 is 1024 x 1024 pixels, which is ideal for sharing on the web but might fall short for print purposes.
Upscaling the images on Midjourney increases the resolution to 2048 x 2048 pixels, which is an improvement but may still be insufficient for certain use cases. However, this upscaling enhances the image quality, making the small details more intricate.
Let’s take a look:
To upscale an image - choose one of the Upscaling options in the side menu:
Choose ‘subtle’ if you like everything about the image and want to make it more detailed. Choose ‘creative’ if you want Midjourney to have a bit more freedom over the upscale.
The upscale makes the image twice as large and adds a bit more detail to the mix in the process.
If you need further upscaling, I recommend using either Gigapixel from Topaz.AI or Magnific.
Improved quality parameter
You can further increase the quality of your generations by using the improved --q 2 parameter. This parameter directs the model to allocate more processing power to the image generation, which can lead to more detailed and refined results, though it may take a bit longer to complete.
Using this parameter will not always guarantee a better outcome, as it depends on how well other aspects of your prompt are dialed in. However, it’s worth experimenting with to see if it enhances the quality of your images, particularly when you’re aiming for greater detail.
Personalization
The personalized model was also updated together with this update and it should be better overall but the biggest change is the addition of rolling personalization codes, meaning we can now save different states of personalization.
You can access a list of all your previous personalization codes on Discord by using /list_personalize_codes command:
Here’s how my personalization has evolved over time:
Prompt: Portrait of a man — ar 4:5 — seed 652506190 — p {ke833gp, yhdh5ku, 6ektuei, no7gtv8, m7yczex, 3wpud8j, vbgt6qu, 5isd2q5, gyekeam} — v 6.1
I’ve covered Personalization more in-depth in an earlier article: read here!
Closing thoughts
While V6.1 is similar to V6, the improvements are noticeable and open up more possibilities for integrating Midjourney AI into various day-to-day tasks.
The enhanced image quality, better text accuracy, faster generation times, and improved upscaling all contribute to a more refined and versatile tool. The advancements in generating finer detailed textures, improved creation of hands, and overall better comprehension showcase the model’s ability to produce more realistic results
Furthermore, it’s a glimpse into the future, giving us a taste of what to expect when V7, with new training data and enhancements, is released. It also hints at how personalization could become a crucial part of image generation.