Midjourney V6 Model Got Major Updates

It’s been only a few weeks since Midjourney released its latest and most advanced AI image model, V6. My initial review of V6 was largely positive, noting the significantly improved image quality and the addition of a new text feature.

Just when many of us thought they had reached a peak, Midjourney surprised us again.

A few days later, Midjourney rolled out an update to V6 with improved image quality, improved text rendering, and more.

In this article, let’s explore the new updates to Midjourney V6.

What’s new in V6?

• Upscales are now twice as fast, utilizing half the GPU resources.
• Notable improvements in aesthetics, coherence, prompt adherence, and overall image quality.
• Enhanced text rendering (remember to enclose text in “quotations” in your prompts).
• Improved performance at high --stylize values
• Upgraded upscaler for reduced artifacts, addressing issues like blur and noise.
• Addition of an AI moderator on the explore page, ensuring safer and cleaner content.

I will be comparing the results generated with the initial V6 model and the latest and updated one. I will also be evaluating the updated text rendering capability of the new V6.

Let’s get started.

Improved images

To gauge the extent of quality improvement, I revisited some prompts I had previously generated with the older V6 model and reprocessed them in the updated V6 model.

Prompt: face made of newspaper, in tabloid art style — ar 4:5

Honestly, it’s challenging to assert that the new model has made strides in image quality. Some users might even prefer the output from the older model. What’s your take?

Let’s do another one.

Prompt: A beautiful gorgeous girl, pale skin and freckles, swimming in clouds, relaxed, meditating in clouds with closed eyes, numb, dreamy, prismatic image, chromatic aberration, pearlescent colors and reflections, golden — ar 4:5

I don’t know if it’s just me, but the older model’s output appeals to me more than the newer version. What’s going on in the new V6?

To be clear, I did a multiple re-generation with the new V6 but it always seem to go for a softer tones and less detailed images than the older model.

Improved text rendering

Now, let’s examine the improvements in text rendering.

Prompt: A large banner for a community book fair, reading “Annual Book Lovers Fair — Discover Your Next Adventure!” The text is surrounded by images of books, reading glasses, and a cozy reading nook.

The results suggest that text rendering for longer texts still requires refinement. Surprisingly, the older model managed this prompt more effectively. The older model even did a better job with this prompt.

Prompt: A promotional card for “Bean Bliss Coffee Shop” featuring a steaming cup of coffee on a rustic table

For shorter texts, the new model shows notable improvement and successfully rendered the requested words. That’s great.

It is important to note that the current V6 model is still in alpha phase. We can expect further changes and enhancements in the coming weeks as the team moves towards the beta phase.

Please note: We’ll continue to push improvements over the next few weeks as we move towards the official beta phase of V6

How to enable V6 model

If you are a Midjourney user, the default image model on Discord is V5.2 instead of V6. To enable V6, open the settings page by using the /settings command.

Select the “Midjourney Model V6 [ALPHA]” and hit the enter key.

Final Thoughts

Midjourney has come a long way since first being introduced in 2021. With each update, the quality and coherence of the images continue to impress.

In fact, the quality of the images is way better than the other giant competitors, like OpenAI’s Dall-E 3 and Stability AI’s SDXL.

And, oh, in case you didn’t know it yet... They are soon to integrate the image generation capability into the Midjourney website.

I will be eagerly following along to see what developments come next for Midjourney.