For some time, I’ve felt just like the AI picture era house is changing into too stagnant. There weren’t any important developments after DALL-E 3, and newer fashions have left me dissatisfied or feeling a bit impartial.

One factor stored me going for months: the discharge of Midjourney V6.

And when it lastly got here, it got here as a sudden Christmas current. Nobody actually anticipated V6 earlier than 2024, so this early sneak peek of its base mannequin — even with out a few of its functionalities — was a welcome shock.

It has been a few days and I am now prepared to present my preliminary ideas on this new mannequin. On this article, I am going to define each enchancment and alter that got here with V6.

What’s New With V6?

Midjourney V6 is the AI picture generator’s newest mannequin, increasing upon the capabilities of earlier iterations and altering a few of its core functionalities. Like with V5, this mannequin isn’t the ultimate model of V6, quite simply the bottom mannequin which might be steadily fine-tuned within the subsequent months.

Nevertheless, it’s nonetheless probably the most succesful Midjourney mannequin thus far, with additions equivalent to:

• Improved output creativity.
• Higher immediate comprehension.
• Strong upscalers.
• Textual content era.

V6 can now take lengthy and complicated prompts, and create correct photographs. It comes at a value although, with loyal Midjourney customers being requested to relearn the best way to immediate for the reason that mannequin is extraordinarily delicate. I’m having the identical subject myself however, on the finish of the day, it’s a small value to pay.

How Are They Bettering The Mannequin?

Aside from increasing its coaching set, Midjourney can also be gathering consumer opinion to enhance its mannequin by means of A/B Testing. That is voluntary however customers can get free hours in the event that they take part, driving extra customers to charge photographs on their web site.

Midjourney V6 Output High quality

Now that you simply’ve been launched to the enhancements Midjourney made on its mannequin, it’s time to see it in motion. Listed below are some output examples from V6, grouped neatly by picture class:

Realism (Portraits)

Immediate A: a younger lady attending a music competition, backlighting, portrait
Immediate B: a physics professor instructing his class, academia, close-up

For those who’ve been following together with my Midjourney evaluations, you realize that I’ve lengthy been annoyed about its tendency to create waxy faces and overemphasize sure options. With V6, that’s turn out to be a factor of the previous.

Each of those photographs look extremely actual. Even in the event you look carefully, there are not any clear indicators that these are generated with an AI in any respect. These examples actually converse to how far Midjourney has come from V5.2 in simply a few months.

Private Rating: 5 out of 5

Realism (Panorama)

Immediate A: solitary stone cabin in an unlimited alpine meadow, wildflowers in bloom, snow-capped peaks within the distance
Immediate B: misty autumn forest path, fallen leaves carpeting the bottom, daylight filtering by means of the bushes at dawn

That is one other excellent rating for me. I attempted to journey Midjourney up through the use of conversational language however its improved coherence allowed it to meet each phrase in my immediate. As for the standard, they’re vibrant and vivid with out going overboard, present no signal of rendering points, the shadows make sense, and the depth of area is constant.

Private Rating: 5 out of 5

3D Renders

Immediate A: a minimap diorama of a quiet stylish library adorned with indoor vegetation
Immediate B: industrial images, a handcrafted ceramic bowl, earth tones, delicate lighting, vegetation

The extra that I take advantage of Midjourney V6, the extra I’m satisfied that it has no weak factors. These are each extremely correct 3D renders of the immediate topics. I significantly just like the composition of the bowl shot, with the pure gentle coming from the window. Then again, the diorama is so detailed however it doesn’t lose that miniature feeling to it.

Private Rating: 5 out of 5

Pastiche

Immediate A: Magneto in a dvd display screen seize of Dragon Ball Z, drawn by Akira Toriyama, animated by Toei animation studio, 1985 Japanese anime
Immediate B: rows upon rows of lavender stretch to the horizon underneath a full moon, within the fashion of vincent van gogh

I’ve by no means actually had any subject with Midjourney imitating different artists earlier than, however they’ve undoubtedly stepped up their sport in V6. Their most noticeable enchancment is subtlety. For instance, after I generated Van Gogh photographs earlier than, it did so by copying Starry Night time carefully which resulted in numerous spirals and stars within the sky. In V6, it took probably the most recognizable traits of each Van Gogh portray and created an approximation of how the mannequin thinks Van Gogh would paint the immediate.

Private Rating: 5 out of 5

Structure and Inside Design

Immediate A: inside, a shed purposed as an artwork studio, bohemian, cottagecore, pure gentle, whimsy, biophilic
Immediate B: exterior, a cathedral by Antoni Gaudí throughout sundown, structure

The inside design picture is just about excellent, in my view. The structure shot, then again, is fairly good itself however the intricacy of the topic led to some rendering points. It’s hidden from afar however in the event you zoom in, you possibly can see some spires bleeding into one another.

Private Rating: 4.5 out of 5

Textual content Era

Immediate A: a bohemian espresso store named “Nook Espresso”
Immediate B: a professor writing “The Idea of Relativity” in a blackboard

Textual content era continues to be a weak spot for AI picture turbines, even with V6. Nevertheless, it’s value noting that this new mannequin could be one of the best in its section for textual content. The nook espresso textual content seems to be somewhat funky, however it’s nonetheless readable for probably the most half. In the meantime, the textual content on the blackboard has some errors, however you possibly can nonetheless see what it’s attempting to jot down.

In my testing, Midjourney V6 has been unbelievable with quick texts (1-3 phrases) however it turns into unreadable past that.

Private Rating: 4 out of 5

Excessive Context

Immediate A: a wide ranging and cinematic portrait of a lone astronaut gazing out on the swirling nebulas of the Horsehead Nebula, their helmet reflecting the cosmic spectacle, as their massive spaceship explodes behind them. delicate and dramatic lighting. evoking a way of awe, marvel, and hazard.
Immediate B: a hyper-realistic portrait of an aged lady, her face etched with the traces of time and expertise, however her eyes shining with knowledge and heat. she sits in a sunlit room, surrounded by mementos of a life well-lived. the portrait captures each the great thing about age and the enduring power of the human spirit. wide-angle. impressed by rembrandt. 

Midjourney isn’t pretty much as good as DALL-E 3 with GPT-4 relating to immediate coherence, however it’s undoubtedly up there. It missed some traces in each prompts, just like the exploding spaceship and mementos, however many of the components are nonetheless current, which is greater than I may say for Midjourney V5.2.

Private Rating: 4 out of 5

Common Rating

When tallied, my common rating of Midjourney V6 is 4.64 out of 5. That’s lower than half a degree away from an ideal rating, which exhibits how unbelievable Midjourney is at its present stage.

If you need extra examples of Midjourney V6’s output, I extremely recommend that you simply learn our comparability articles towards V5 and different AI picture turbines.

Execs & Cons of Utilizing Midjourney V6

Midjourney V6 vs. Different AI Picture Mills

DALL-E 3

Launched in October 2023, DALL-E 3 is the third model of OpenAI’s picture generator. Like V6, it was a major evolution from its earlier iteration, with a deal with each comprehension and textual content era. It’s obtainable by means of ChatGPT Plus or with Bing Create.

High quality Comparability

Portraits

Panorama

3D Product Mockups

Textual content Era

Excessive Context Prompts

What Makes DALL-E 3 Higher Than Midjourney V6?

• Nonetheless considerably higher at nuance.
• It may be accessed by means of a browser.
• Much less vulnerable to AI hallucination and rendering points.
• Quicker era time than the present Midjourney mannequin.
• GPT-4 processes your conversations or prompts into ones that may be higher understood by DALL-E 3.

What Makes DALL-E 3 Worse Than Midjourney V6?

• Midjourney can now do textual content higher than DALL-E 3.
• Midjourney is best at each realism and digital artwork.
• It doesn’t have the identical customization options as Midjourney.
• You need to use artist names as prompts for Midjourney.
• DALL-E doesn’t provide you with management over the output’s facet ratio.

Meta

Meta’s AI picture generator is a text-to-image generative mannequin which makes use of a mannequin referred to as Emu. It’s utterly free however it’s additionally morally ambiguous, extra so than different picture turbines, as this mannequin makes use of information from Fb and Instagram customers as its coaching set.

High quality Comparability

Portraits

Panorama

3D Product Mockups

Textual content Era

Excessive Context Prompts

What Makes Meta Higher Than Midjourney V6?

• Considerably sooner era velocity.
• Meta is free.

What Makes Meta Worse Than Midjourney V6?

• It doesn’t save previous prompts and paintings.
• It doesn’t have any customization options.
• Meta’s creativity isn’t pretty much as good as Midjourney.
• Meta can’t do textual content and doesn’t observe lengthy prompts properly.
• Meta makes use of Fb and Instagram consumer information as its coaching set.

Wrapping Up

It’s somewhat too early to inform, but when that is how good Midjourney V6 already is even at its base mannequin, then I don’t see any level in investing in different AI picture turbines. It’s so good that it blows different fashions out of the water. Solely DALL-E can catch as much as Midjourney now, they usually’re not even remotely shut.

That stated, it nonetheless has a pair shortcomings, significantly in comprehension and lengthy textual content era. However then once more, so does each different AI generator. 

In some unspecified time in the future, a mannequin will attain the purpose of singularity in AI picture era, and we’ll be shifting on to newer frontiers like text-to-video or image-to-video. I actually imagine that Midjourney’s going to be on the pinnacle of AI picture era — the one that can usher this inventive future.