Do you know this feeling when you look at a picture and immediately know—it's Midjourney? V6 is yet another step towards the time when this feeling will become rare.
Tim Walker's haute-couture frontal portrait of clear white ethereal android with translucent skin drowning in sea of wires. Red and cyan hues, glowing highlights, dark shadows --v 6.0
Tim Walker's haute-couture frontal portrait of clear white ethereal android with translucent skin drowning in sea of wires. Red and cyan hues, glowing highlights, dark shadows --v 5.2
In Part 1 of this study, we look in-depth at Midjourney’s newly released model, its strengths, weaknesses, and key changes from V5.2.
MRI scan of robot samurai --v 6.0
MRI scan of robot samurai --v 5.2
Japanese fox god of winter death and rebirth --v 6.0
Japanese fox god of winter death and rebirth --v 5.2
intricate raven god --v 6.0
intricate raven god --v 5.2
Quick Facts About V6
The current V6 is an Alpha test; thus, things may change.
V6 is more accurate in following a prompt (and is better with longer prompts).
The new model is more coherent.
It has improved image-prompting capabilities.
There are two new upscalers, with Subtle and Creative modes (both increase resolution by 2×).
There is an “unopinionated” --style raw mode (similar to V5.2).
You can add text to your images now!
Updated: the Pan, Zoom, Vary (Region) work now (and the in-depth guide on those is coming soon!), while /tune, and the new V6 /describe aren't available. But they are announced to arrive in the nearest weeks.
But apart from these lines—what is V6?
Hyperrealism
Every new version of Midjourney comes with a boost in photorealism, and V6 is no exception. In fact, the keyword that defines the new model best is Hyperrealism.
Rinko Kawauchi's photographic portrait of girl space pilot --v 6.0 --style raw
Rinko Kawauchi's photographic portrait of girl space pilot --v 5.2 --style raw
To summarize briefly, the level of photorealism in V6, especially in photographic styles, is mind-bending.
steampunk traveler monk in desert by Lynsey Addario --v 6.0
steampunk traveler monk in desert by Lynsey Addario --v 5.2
Brett Walker's photographic portrait of Daniel Defoe --v 6.0 --style raw
Brett Walker's photographic portrait of Daniel Defoe --v 5.2 --style raw
Kourtney Roy's photographic portrait of old seafarer --v 6.0
Kourtney Roy's photographic portrait of old seafarer --v 5.2
What makes photorealistic images in Midjourney V6 look so amazing are the imperfections: lens aberrations, intentionally over-highlighted areas, accidental out-of-focus elements, and various film effects (which we will dive into in the 'Details' chapter).
Serge Lutens's photographic portrait of young pirate-queen --v 6.0 --style raw
Serge Lutens's photographic portrait of young pirate-queen --v 5.2 --style raw
Mitsuo Katsui's photograph of Hatsune Miku --stylize 175 --v 6.0
Mitsuo Katsui's photograph of Hatsune Miku --stylize 175 --v 5.2
last samurai. Cinematic portrait by Wong Kar-Wai --v 6.0 --style raw
last samurai. Cinematic portrait by Wong Kar-Wai --v 5.2 --style raw
And, of course, it's not just about portraits...
gigantic floating fortress in Northern Sea port. Photograph by Allan Sekula --v 6.0
gigantic floating fortress in Northern Sea port. Photograph by Allan Sekula --v 5.2
time-lapse photography over Tbilisi --stylize 175 --v 6.0
time-lapse photography over Tbilisi --stylize 175 --v 5.2
top view haute-couture advertising food photograph for Japanese restaurant with molecular cuisine. Minimal composition, cyan and red hues --v 6.0
top view haute-couture advertising food photograph for Japanese restaurant with molecular cuisine. Minimal composition, cyan and red hues --v 5.2
In some cases, however, where a more subtle look-and-feel would be preferable, V6's hyperrealism—with a tendency to oversharp things—may be considered an overkill.
16th century knight. Official portrait by Richard Mosse --v 6.0
16th century knight. Official portrait by Richard Mosse --v 5.2
Martin Schoeller's close-up portrait of frontman of death-metal band --v 6.0
Martin Schoeller's close-up portrait of frontman of death-metal band --v 5.2
Ara Guler's portrait of 1970s Istanbul casino gambler --v 6.0
Ara Guler's portrait of 1970s Istanbul casino gambler --v 5.2
V6 indeed marks a significant milestone in the evolution of ultra-realistic AI art. The Midjourney team has—once again!—surpassed expectations, blurring the line between real photographs and AI-generated imagery like never before. Which is both thrilling and a bit unsettling. (-’๏_๏’-)
Detail Insanity
Every time before the release of Midjourney’s next model, I hold my breath for what they will do with details.
Rinko Kawauchi's photographic portrait of girl space pilot --v 6.0 --style raw
Rinko Kawauchi's photographic portrait of girl space pilot --v 5.2 --style raw
At the risk of repeating myself: the level of details in V6 is, for the lack of a better word, insane. And the new upscalers take it even further (more on them in Part 4 of this deep dive).
Carlo Crivelli's painting depicting red knight --v 6.0 --style raw
Carlo Crivelli's painting depicting red knight --v 5.2 --style raw
.jpg)
Leon Bakst's illustration for Darth Vader's ballet sceni ostume --v 6.0
Leon Bakst's illustration for Darth Vader's ballet scenic costume --v 5.2
Ivan Bilibin's painting depicting lord of winter riding white wolf --stylize 275 --v 6.0
Ivan Bilibin's painting depicting lord of winter riding white wolf --stylize 275 --v 5.2
But while the intricacy of the images went up steadily with each new model, some things were lost along the way, namely, the textures. The refinement of V4 dialed up tenfold in V5+ made it almost impossible to achieve effects like film grain or true brushstrokes, for instance.
.jpg)
street photograph by Miroslav Tichy --v 6.0 --style raw
street photograph by Miroslav Tichy --v 5.2 --style raw
broad brushstrokes painted portrait of Troll Hunter in Dutch Golden Age style --v 6.0
broad brushstrokes painted portrait of Troll Hunter in Dutch Golden Age style --v 5.2
pixelated screenshot of 1980s PC game. Underworld level --v 6.0
pixelated screenshot of 1980s PC game. Underworld level --v 5.2
And while true grain still seems out of reach, V6 is a definite step towards that lost rawness.The textures are amazing, and the new model shines against V5, where “non-refined” visual styles are required.
cross processing print of flowers in wind --v 6.0
cross processing print of flowers in wind --v 5.2
dreamy double-exposure portrait of Amelia Erhart. Planes overlay --stylize 55 --v 6.0
dreamy double-exposure portrait of Amelia Erhart. Planes overlay --stylize 55 --v 5.2
dreamy girl by Marianna Rothen --v 6.0 --style raw
dreamy girl by Marianna Rothen --v 5.2 --style raw
Shifted Composition
Another concept that describes V6 well is unconventional composition, a significant shift from the more structured and balanced approach of V5.2.
interdimensional arcane beast by Tyrus Wong --stylize 175 --v 6.0 --style raw
interdimensional arcane beast by Tyrus Wong --stylize 175 --v 5.2 --style raw
Whereas previous models seek geometrical perfection, golden ratio, and central subjects, V6 strives for asymmetry and often moves its main subjects away from the middle of the frame.
Runner. Motion blur movement --v 6.0 --style raw
Runner. Motion blur movement --v 5.2 --style raw
Black Panther by Jamel Shabazz --v 6.0
Black Panther by Jamel Shabazz --v 5.2
lonely post-apocalyptic ranger by Corey Arnold --v 6.0
lonely post-apocalyptic ranger by Corey Arnold --v 5.2
This may offer more dynamic and engaging visuals, but it also requires adapting prompts accordingly. Even a small adjustment can bring back both a central composition and symmetry.
Margaret Bourke-White's photograph of enormous Doomsday device in clouds. Colossal scale --v 6.0
Margaret Bourke-White's photograph of enormous Doomsday device in clouds. Colossal scale --v 5.2
Margaret Bourke-White's photograph of enormous Doomsday megastructure floating in clouds. Colossal scale. Central composition --v 6.0
Margaret Bourke-White's photograph of enormous Doomsday megastructure floating in clouds. Colossal scale. Central composition --v 5.2
Margaret Bourke-White's symmetrical photograph of enormous Doomsday megastructure floating in clouds. Colossal scale. Central composition --v 6.0
Margaret Bourke-White's symmetrical photograph of enormous Doomsday megastructure floating in clouds. Colossal scale. Central composition --v 5.2
Although, at times, V6's shifts in composition might seem unrequested and unjustified, it is encouraging to see that Midjourney experiments with unconventional compositions, and expanded the overall variability of the output. Speaking of which…
Variability
With V6, one of the first things you'll notice is the increased variability in outcomes from the same prompt.
Silver Warlock by Jessie Willcox Smith --v 6.0
Silver Warlock by Jessie Willcox Smith --v 5.2
It's like the Midjourney developers cranked up the default --chaos dial. As a result, four variations from a single prompt often show more distinct differences than in earlier models.
retrofuturistic home appliance --v 6.0
retrofuturistic home appliance --v 5.2
Op-art depicting Chimera --v 6.0
Op-art depicting Chimera --v 5.2
black flowers by Maria Sibylla Merian --v 6.0
black flowers by Maria Sibylla Merian --v 5.2
This increased diversity is especially noticeable with abstract concepts or prompts that leave space for interpretation, and is also apparent in basic prompts: by [artist name], designed to show how an artistic style works in Midjourney on its own.
by Kazumasa Nagai --v 6.0
by Kazumasa Nagai --v 5.2
This, in many cases, leads to an artistic style representation that is more creative, more interesting, and, sometimes, more faithful towards the diversity of the real-life prototype.
And on the topic of artistic styles…
Artistic Styles In Midjourney V6
Style modifiers, or simply styles, are the names or titles you can reference in your prompts to summon a specific visual flair, technique, genre, subject, or context for your image.
It’s a tradition by now that every new model becomes better at knowing the source material and re-creating it in its output. However, Midjourney never ceases to surprise by how dramatic the change is.
Japanese vintage poster --v 6.0
Japanese vintage poster --v 5.2
by Leonetto Cappiello --v 6.0
by Leonetto Cappiello --v 5.2
Let’s compare how style modifiers work in V6 vs. V5.2 using styles from our catalog and a variety of prompts.
cutout animation scene from Peppa Pig --v 6.0
cutout animation scene from Peppa Pig --v 5.2
dark fantasy photoshoot by Annie Leibovitz --v 6.0 --style raw
dark fantasy photoshoot by Annie Leibovitz --v 5.2 --style raw
Moebius' illustration depicting robot portrait --v 6.0
Moebius' illustration depicting robot portrait --v 5.2
In certain instances, the most significant shift isn't in quality, but in the understanding of the source material. V6 appears to have a slightly different familiarity with some artists' work when compared to the same artists’ interpretations by V5+ models.
Comic-strip style --v 6.0
Comic-strip style --v 5.2
by Luigi Ghirri --v 5.2 --style raw
But however great the styles might be by themselves, what truly turns them from a mere interpretation of the original work to something unique is the prompt that you add to them—e.g., your creative vision, converted into text.
Text!
Remember how we fought Midjourney to make it NOT add text to our pictures? Well, now you can ADD text to your pictures deliberately. Well… to a certain extent. (-‿◦)
.jpg)
hands holding newspaper with heading "Text in Midjourney- real or hoax?" text --stylize 55 --v 6.0 --style raw
.jpg)
hands holding newspaper with heading "Text in Midjourney- real or hoax?" text --stylize 55 --v 5.2 --style raw
The key word here is perseverance. You can get the (almost-)perfect result, but it will be paid in countless re-rolls and updates to your prompt.
text "Midlibrary" written in floral-motives-font --v 6.0
text "Midlibrary" written in floral-motives-font --v 5.2
.jpg)
text "Bond, James Bond". Elegant letters on a minimalist bold graphical 1950s James Bond movie poster --stylize 85 --v 6.0
.jpg)
text "Bond, James Bond". Elegant letters on a minimalist bold graphical 1950s James Bond movie poster --stylize 85 --v 5.2
.jpg)
letters "Fhtagn!" old CGA-screen pixelized text with little pixelated Cthulhu ASCII-pictogram --stylize 75 --v 6.0
.jpg)
letters "Fhtagn!" old CGA-screen pixelized text with little pixelated Cthulhu ASCII-pictogram --stylize 75 --v 5.2
Here are a few important things to remember when working with text in Midjourney V6:
Put your text in “quotations.”
Leveraging --stylize values (lower might be better), and switching to --style raw can improve the result noticeably.
Expectedly, shorter words work better. However, with enough persistence you can make Midjourney produce longer words and even whole phrases.
Adding words like text, letters, etc. to your prompt may improve the outcome.
Sometimes, you might want to double the presence of your target text in a prompt by repeating it twice in different parts of the prompt.
To demonstrate how these reccomendations apply—a few more live examples:
text "Fly, you fools!". Large letters "Fly, you fools!" in trash poster duotone-print collage style --stylize 75 --v 6.0
text "Fly, you fools!". Large letters "Fly, you fools!" in trash poster duotone-print collage style --stylize 75 --v 5.2
text "Follow the white rabbit" graffiti in blacklight paint --stylize 55 --v 6.0
text "Follow the white rabbit" graffiti in blacklight paint --stylize 55 --v 5.2
.jpg)
text "2049". Cyberpunk nixie-tubes font number glowing in the dark --stylize 50 --v 6.0 --style raw
.jpg)
text "2049". Cyberpunk nixie-tubes font number glowing in the dark --stylize 50 --v 5.2 --style raw
Another key element influencing the outcome in V6 is the Aspect Ratio, or --ar, of your image.
Here is how Midjourney tried to squeeze letters into a vertical format in my numerous attempts.
noodle-letters. Text "Ramen" made of noodles --v 6.0
noodle-letters. Text "Ramen" made of noodles --v 5.2
noodle-letters. Text "Ramen" made of noodles --stylize 75 --v 6.0
noodle-letters. Text "Ramen" made of noodles --stylize 75 --v 5.2
noodle-letters. Text "Ramen" made of noodles --stylize 35 --v 6.0
noodle-letters. Text "Ramen" made of noodles --stylize 35 --v 5.2
Yet, the moment I allowed for more horizontal space, the model promptly delivered a more accurate result:
noodle-letters. Text "Ramen" made of noodles --stylize 35 --v 6.0 --style raw
noodle-letters. Text "Ramen" made of noodles --stylize 35 --v 5.2 --style raw
V6 Raw Mode
Raw is an alternative—“unopinionated”—model of Midjourney, existing for both, V6 and V5.2. It activates if you add --style raw to your prompt, and creates an image with ”less automatic beautification applied, which can result in a more accurate match when prompting for specific styles.“
anaglyph --v 6.0 --style raw
The Raw model is more literal, and thus less “creative.” It also tends to lean more towards existing titles when the prompt is ambiguous, i.e., offers multiple interpretations. Like in cases with destiny (also a game), perfect circle (also a music band), and David (also a statue (-‿◦)).
perfect circle --v 6.0 --style raw
destiny --v 6.0 --style raw
David --v 6.0 --style raw
Whereas the default outcome is varied and presents different options, the raw results are consistently focused on reproducing the titles behind the prompts (the game, the music band's frontman, and the statue).
One other difference between the two versions is that the default V6 offers more variability in its results for each prompt, whereas the Raw model’s options are much closer to each other.
imagination --v 6.0 --style raw
And what about more-than-one-word prompts? How does switching from the default to the raw model affect the outcome then?
colored bas-relief depicting alien flower growing though concrete brutalist buildings --v 6.0
colored bas-relief depicting alien flower growing though concrete brutalist buildings --v 6.0 --style raw
elaborate electromagnetic quantum collider mechanism. Inside details, technical labels and markings, complex electronics, mechanics, engineering schematics overlay --v 6.0
elaborate electromagnetic quantum collider mechanism. Inside details, technical labels, markings, complex electronics, mechanics, engineering schematics overlay --v 6.0 --style raw
weather map --v 6.0 --style raw
Finally, let's experiment with how styles/style modifiers work in both modes.
ethereal star beast. Mike Mignola's illustration --v 6.0
ethereal star beast. Mike Mignola's illustration --v 6.0 --style raw
David Uhl's painting depicting Amelia Earhart --v 6.0
David Uhl's painting depicting Amelia Earhart --v 6.0 --style raw
.jpg)
Eugenio Recuenco's photograph of Thom Yorke --v 6.0
Eugenio Recuenco's photograph of Thom Yorke --v 6.0 --style raw
In conclusion, the --style raw parameter in Midjourney is an effective tool for diversifying results, strengthening styles that imply photorealism, or pushing the AI towards a more literal interpretation of your prompt.
In Part 2 of this exploration we will learn to talk to the new Midjourney, see what prompting strategies work best, and how complex our prompts can get. We will also test the new Image Prompting capabilities, observe how Image Weight affects the outcome, and how Blending works in V6. Continue reading →
In-depth Guide to Midjourney V6
You can help us maintain and expand Midlibrary and produce more regular educational content of higher quality. And keep it free for all!
Support Midlibrary on Patreon! →
Midjourney Style Roulette
ⓘ New styles on every page!
/explore Midjourney styles
