Gold Penguin Logo with Text

Midjourney V6 is Incredible – My Initial Thoughts and Prompts

Midjourney recently unveiled its newest version: Midjourney V6. It's their most advanced model by far, it's incredible! Here's what you should expect when using this new model.
Updated December 26, 2023
A lone ship in the middle of the ocean, generated with Midjourney
A lone ship in the middle of the ocean, generated with Midjourney

For a while, I've felt like the AI image generation space is becoming too stagnant. There weren't any significant developments after DALL-E 3, and newer models have left me disappointed or feeling a bit neutral.

One thing kept me going for months: the release of Midjourney V6.

And when it finally came, it came as a sudden Christmas present. No one really expected V6 before 2024, so this early sneak peek of its base model — even without some of its functionalities — was a welcome surprise.

It's been a couple of days and I'm now ready to give my initial thoughts on this new model. In this article, I'll outline every improvement and change that came with V6.

What's New With V6?

Midjourney V6 is the AI image generator’s latest model, expanding upon the capabilities of previous iterations and changing some of its core functionalities. Like with V5, this model isn’t the final version of V6, rather just the base model which will be gradually fine-tuned in the next months.

However, it’s still the most capable Midjourney model to date, with additions such as:

  • Improved output creativity.
  • Better prompt comprehension.
  • Robust upscalers.
  • Text generation.

V6 can now take long and complex prompts, and create accurate images. It comes at a cost though, with loyal Midjourney users being asked to relearn how to prompt since the model is extremely sensitive. I’m having the same issue myself but, at the end of the day, it’s a small price to pay.

How Are They Improving The Model?

Apart from expanding its training set, Midjourney is also gathering user opinion to improve its model through A/B Testing. This is voluntary but users can get free hours if they participate, driving more users to rate images on their website.

Midjourney V6 Output Quality

Now that you’ve been introduced to the improvements Midjourney made on its model, it’s time to see it in action. Here are some output examples from V6, grouped neatly by image category:

Realism (Portraits)

Prompt A: a young woman attending a music festival, backlighting, portrait
Prompt B: a physics professor teaching his class, academia, close-up

If you’ve been following along with my Midjourney reviews, you know that I’ve long been frustrated about its tendency to create waxy faces and overemphasize certain features. With V6, that’s become a thing of the past.

Both of these images look incredibly real. Even if you look closely, there are no clear indicators that these are generated with an AI at all. These examples really speak to how far Midjourney has come from V5.2 in just a couple of months.

Personal Score: 5 out of 5

Realism (Landscape)

Prompt A: solitary stone cabin in a vast alpine meadow, wildflowers in bloom, snow-capped peaks in the distance
Prompt B: misty autumn forest path, fallen leaves carpeting the ground, sunlight filtering through the trees at daybreak

This is another perfect score for me. I tried to trip Midjourney up by using conversational language but its improved coherence allowed it to fulfill every word in my prompt. As for the quality, they’re bright and vivid without going overboard, show no sign of rendering issues, the shadows make sense, and the depth of field is consistent.

Personal Score: 5 out of 5

3D Renders

Prompt A: a minimap diorama of a quiet chic library adorned with indoor plants
Prompt B: commercial photography, a handcrafted ceramic bowl, earth tones, soft lighting, plants

The more that I use Midjourney V6, the more I’m convinced that it has no weak points. These are both incredibly accurate 3D renders of the prompt subjects. I particularly like the composition of the bowl shot, with the natural light coming from the window. On the other hand, the diorama is so detailed but it doesn’t lose that miniature feeling to it.

Personal Score: 5 out of 5

Pastiche

Prompt A: Magneto in a dvd screen grab of Dragon Ball Z, drawn by Akira Toriyama, animated by Toei animation studio, 1985 Japanese anime
Prompt B: rows upon rows of lavender stretch to the horizon under a full moon, in the style of vincent van gogh

I’ve never really had any issue with Midjourney imitating other artists before, but they’ve definitely stepped up their game in V6. Their most noticeable improvement is subtlety. For example, when I generated Van Gogh images before, it did so by copying Starry Night closely which resulted in a lot of spirals and stars in the sky. In V6, it took the most recognizable characteristics of every Van Gogh painting and created an approximation of how the model thinks Van Gogh would paint the prompt.

Personal Score: 5 out of 5

Architecture and Interior Design

Prompt A: interior, a shed purposed as an art studio, bohemian, cottagecore, natural light, whimsy, biophilic
Prompt B: exterior, a cathedral by Antoni Gaudí during sunset, architecture

The interior design image is pretty much perfect, in my opinion. The architecture shot, on the other hand, is pretty good itself but the intricacy of the subject led to some rendering issues. It’s hidden from afar but if you zoom in, you can see some spires bleeding into each other.

Personal Score: 4.5 out of 5

Text Generation

Prompt A: a bohemian coffee shop named "Corner Coffee"
Prompt B: a professor writing "The Theory of Relativity" in a blackboard

Text generation continues to be a weak spot for AI image generators, even with V6. However, it’s worth noting that this new model might be the best in its segment for text. The corner coffee text looks a little funky, but it’s still readable for the most part. Meanwhile, the text on the blackboard has some mistakes, but you can still see what it’s trying to write.

In my testing, Midjourney V6 has been incredible with short texts (1-3 words) but it becomes unreadable beyond that.

Personal Score: 4 out of 5

High Context

Prompt A: a breathtaking and cinematic portrait of a lone astronaut gazing out at the swirling nebulas of the Horsehead Nebula, their helmet reflecting the cosmic spectacle, as their large spaceship explodes behind them. soft and dramatic lighting. evoking a sense of awe, wonder, and danger.
Prompt B: a hyper-realistic portrait of an elderly woman, her face etched with the lines of time and experience, but her eyes shining with wisdom and warmth. she sits in a sunlit room, surrounded by mementos of a life well-lived. the portrait captures both the beauty of age and the enduring strength of the human spirit. wide-angle. inspired by rembrandt. 

Midjourney isn’t as good as DALL-E 3 with GPT-4 when it comes to prompt coherence, but it’s definitely up there. It missed some lines in both prompts, like the exploding spaceship and mementos, but most of the elements are still present, which is more than I could say for Midjourney V5.2.

Personal Score: 4 out of 5

Average Score

When tallied, my average score of Midjourney V6 is 4.64 out of 5. That’s less than half a point away from a perfect score, which shows how incredible Midjourney is at its current stage.

If you want more examples of Midjourney V6’s output, I highly suggest that you read our comparison articles against V5 and other AI image generators.

Pros & Cons of Using Midjourney V6

PROS

CONS

  • Significant model improvement across the board.

  • Amazing at prompt comprehension and text generation.

  • Can now do realistic photographs.

  • Ethically outsources data intake to its users through A/B testing.

  • The best AI image generator in the market.

  • Slow generation speed at its current phase.

  • Removed some of V5.2’s features like region variations and panning. (These will return later)

  • Despite the new model, Midjourney still has no separate web application.

Midjourney V6 vs. Other AI Image Generators

DALL-E 3

Released in October 2023, DALL-E 3 is the third version of OpenAI’s image generator. Like V6, it was a significant evolution from its previous iteration, with a focus on both comprehension and text generation. It’s available through ChatGPT Plus or with Bing Create.

Quality Comparison

Portraits

Landscape

3D Product Mockups

Text Generation

High Context Prompts

What Makes DALL-E 3 Better Than Midjourney V6?

  • Still significantly better at nuance.
  • It can be accessed through a browser.
  • Less prone to AI hallucination and rendering issues.
  • Faster generation time than the current Midjourney model.
  • GPT-4 processes your conversations or prompts into ones that can be better understood by DALL-E 3.

What Makes DALL-E 3 Worse Than Midjourney V6?

  • Midjourney can now do text better than DALL-E 3.
  • Midjourney is better at both realism and digital art.
  • It doesn’t have the same customization features as Midjourney.
  • You can use artist names as prompts for Midjourney.
  • DALL-E doesn’t give you control over the output’s aspect ratio.

Meta

Meta’s AI image generator is a text-to-image generative model which uses a model called Emu. It’s completely free but it’s also morally ambiguous, more so than other image generators, as this model uses data from Facebook and Instagram users as its training set.

Quality Comparison

Portraits

Landscape

3D Product Mockups

Text Generation

High Context Prompts

What Makes Meta Better Than Midjourney V6?

  • Significantly faster generation speed.
  • Meta is free.

What Makes Meta Worse Than Midjourney V6?

  • It doesn’t save past prompts and artwork.
  • It doesn’t have any customization features.
  • Meta’s creativity isn’t as good as Midjourney.
  • Meta can’t do text and doesn’t follow long prompts well.
  • Meta uses Facebook and Instagram user data as its training set.

Wrapping Up

It’s a little too early to tell, but if this is how good Midjourney V6 already is even at its base model, then I don’t see any point in investing in other AI image generators. It’s so good that it blows other models out of the water. Only DALL-E can catch up to Midjourney now, and they're not even remotely close.

That said, it still has a couple shortcomings, particularly in comprehension and long text generation. But then again, so does every other AI generator. 

At some point, a model will reach the point of singularity in AI image generation, and we'll be moving on to newer frontiers like text-to-video or image-to-video. I truly believe that Midjourney's going to be at the pinnacle of AI image generation — the one that will usher this creative future.

Want To Learn Even More?
If you enjoyed this article, subscribe to our free monthly newsletter
where we share tips & tricks on how to use tech & AI to grow and optimize your business, career, and life.
Written by John Angelo Yap
Hi, I'm Angelo. I'm currently an undergraduate student studying Software Engineering. Now, you might be wondering, what is a computer science student doing writing for Gold Penguin? I took up studying computer science because it was practical and because I was good at it. But, if I had the chance, I'd be writing for a career. Building worlds and adjectivizing nouns for no other reason other than they sound good. And that's why I'm here.
Subscribe
Notify of
guest

1 Comment
Most Voted
Newest Oldest
Inline Feedbacks
View all comments
Join Our Newsletter!
If you enjoyed this article, subscribe to our free monthly newsletter where we share tips & tricks on how to use tech & AI to grow and optimize your business, career, and life.
magnifiercross