Midjourney V6 was teased throughout 2023, and finally released on December 22, 2023.
This has the longest wait so far between all Midjourney releases.
I've been playing with it a ton and hanging out in the Midjourney discord to see what the community has been doing.
The reception to this model has been overwhelmingly positive in Midjourney communities and on social media:
Amazing how when you see a picture from v5.2 you think "this is already maximum quality, surely v6 can't improve on this?" - and then you see obvious improvements from v6.
Block-Rockig-Beats
Note that is an alpha release, and the final version of Midjourney V6 may be different.
What's new
- More accurate prompting, and can follow longer prompts more accurately
- Improved image coherence
- Improved image prompting and remix
- Ability to include text, by wrapping it in quotations: "example"
- Improved upscaling
My first look at V6
One point that Holz stresses in the announcement is that prompting with Midjourney V6 is significantly different than prompting in Midjourney V5, and that you will need to relearn how to prompt.
He advises to get rid of junk like "8k, best quality", and be very descriptive about what you want, because the V6 model understands prompts much better than previous models.
I love this. It means in the future we'll rely less and less on prompt engineering "tricks" and more and more on just describing what exactly we want.
The first thing I did was pull up some prompts from my V5 article and rewrite them to what I thought would be suitable for V6.
Old Prompting Style
21 year old chinese girl, in school, beauty, long black hair, sun light, insane detail, smooth light, real photography fujifilm superia, full HD, taken on a Canon EOS R5 F1.2 ISO100 35MM --ar 4:3 --s 750 --v 6.0This is a common prompting style you will see people use with V5 and previous models. Lots of qualifiers like "insane detail" and lots of camera related words.
New Prompting Style
real photography of 21 year old chinese girl in sunlight --ar 4:3 --s 750 --v 6.0This is the same prompt stripped of all the "clutter". You can see nothing is lost by stripping out all of the camera-words and "quality words".
I've upscaled one image. The quality is unreal:
# of objects in the same prompt
In my testing, I've found that V6 can consistently handle about 5 specific objects in a single image. As I added more than that, Midjourney started dropping elements.
That's not to say 5 is the maximum; far from it! It's just that 5 and under objects is where you can get very consistent accuracy.
5 objects
Objects are all accurate and image is cohesive.
cat wearing sunglasses next to lemon and camera, with a red lamp
6 objects
Still performs very well, but there is a slight loss of accuracy.
cat wearing sunglasses next to lemon and camera and nutcracker, with a red lamp
13 objects
For fun, I tried many objects. Even though some objects are dropped, I'm VERY impressed by the results:
cat wearing sunglasses and top hat next to lemon and camera and nutcracker and watermelon and roast duck on a plate and a saltshaker and a donut, with a red lamp, balloons
Let's get into the prompts.
Realistic People
You don't have to prompt very much to get very realistic result; it's the model default.
- V6 is by far the most realistic model to date
- Add the
--style rawparameter for a more photographic / less opinionated / more literal result - Setting a very high
--stylizevalue (default 100) will create more aesthetic results (whether this means a more realistic image is up for debate)
a woman with long blond hair and pink nails, in the style of light black and light beige, cluj school, expressive eyes, webcam, soft edges, allyson grey, dreamlike quality
White background, tied hair, makeup hair, reduce broken hair, black long hair, magazine feeling, clean and refreshing, adding makeup magazine feeling, rinko kawauchi style
A pretty young woman sits by the misty lake at dusk, looking at the camera, wearing iridescent satin under the glaring red sun, ethereal light, masterpiece cinematography --ar 9:16
girl playing in the water on a beach --style raw --stylize 90 --ar 9:16
Extreme close-up photo. The face of a young Arabic woman with long hair and hazel eyes, expressing a sense of wonder. Misty sunrise with golden light. In the style of fashion professional photo shooting --ar 4:5 --style raw --stylize 250
the joker played by heath ledger and joaquin phoenix with green and pink underglow; vintage 1980s horror movie grainy film footage directed by; park chan-wook, Stanley Kubrick, Guillermo del Toro, kentaro miura --ar 4:5 --style raw --stylize 70
a black and white photo of a man sitting on the floor, studio portraiture, john lurie, george stefanescu, portrait, performance --ar 4:3 
realistic asuka soryu langley in the ruins of a city --ar 2:3 --s 70
realistic portrait photo of real life goku dragon ball Z --ar 2:3 --s 600Lower the --stylize value
I find setting the --stylize value below the default can sometimes make the image more realistic.
Compare these generations:
Low --stylize value
photography of woman with red hair in sunlight --ar 4:3 --seed 997975533 --stylize 50High --stylize value
photography of woman with red hair in sunlight --ar 4:3 --seed 997975533 --v 6.0 --stylize 750Even though the high --stylize generation is more "aesthetic", the imperfections of the low --stylize make it look more realistic and less like a Midjourney generation.
A reddit user has noticed that when prompting "average [country] person", the lowest --s value of 0 gives a much more "average" result:
Camera angles
V6 is much more reponsive to different camera angles.
low-angle photo from below, of a woman wearing sunglasses --ar 16:9If you're not getting the angle you want, or you want to empasize your angle, you can use a slider AKA prompt weight:
Without additional slider
aerial photo from above, of a woman wearing sunglasses --ar 16:9 With additional slider
aerial photo from above, of a woman wearing sunglasses:: aerial photo::0.7 --ar 16:9Photography
Product Photography
mockup of gold face serum, creative direction, product styling, product portfolio , product photography, negative space --style raw --s 90 --ar 2:3
Perfume on the the water. witha magical and luxurious quality, gold and blue color scheme, frozen in place water --ar 9:16Text
One of the bigger changes with V6 is text prompting. You can add quotations around text, like "2024" to generate the text.
- Accuracy is a mixed bag but words better with fewer words.
- Text works better with "
--style raw"
Fireworks dispersing into the shape of "2024" in the night sky,a azure night sky,celebratory crowd in the streets,cartoon style
the cutest animal in the world with the words "happy new year"
A whimsical design of the phrase "Espresso yourself" shaped like steam rising from a hot cup of espresso, on a soft, muted background. Created Using: fluid steam shapes, gentle curves, soothing color palette, whimsical typography, coffee aroma visual, glibatree prompt, hd quality, natural look --v 6Illustration
What's striking about illustration is that you can prompt so many elements now, but in various styles. A lot of my testing was to see how detailed accurate prompts could be generated in certain styles.
anime girl is wearing black and has blond hair, contrasting shadows, dark pink, pulp comics, leather/hide, elegant, emotive faces, stipple, poster art, neo-punk rebellion, light beige face skin, anti-gloss, in the style of pop art iconography, one color face skin, Studio Ghibli style --ar 16:9
In the home office, a kid with a cat and a dog sits in front of the laptop screen. On the laptop you can see a construction drawing. Potted plants all over the room. Many notes are stuck on the walls and connected with strings. Cinematic illustration, Otake Chikuha, irina nordsol kuzmina --ar 16:9
a girl wearing a witch hat and a fluffy bear traveling through a wide expanse of land, there is a field of sunflowers with a mountain village in the distance. swiss mountains in the background. Cinematic illustration, Otake Chikuha, irina nordsol kuzmina --ar 16:9 
female cyborg character design by yoji shinkawa, intricate and elegant, with gold veins that look like wabi sabi throughout --ar 9:16
female ninja with eyepatch, holding 2 swords, flowing robes, scar , art style of yoji shinkawa --ar 2:3 --c 2
female spacepunk knight, a character design by yoji shinkawa and tsutomu nihei, ink and wash, armored intricate and elegant --ar 16:9 --c 2 --s 400
line art with solid coloring, At-At on Hoth in the snow, dynamic action by Moebius and Syd Mead --ar 16:9Traditional Mediums
female cyborg character design by yoji shinkawa, intricate and elegant, with gold veins that look like wabi sabi throughout --ar 9:16female ninja with eyepatch, holding 2 swords, flowing robes, scar , art style of yoji shinkawa --ar 2:3 --c 2Anime
asuka langley anime poster , epic composition cinematic angle --ar 16:9
cell shaded anime style of studio ghibli and makoto shinkai, man in white shirt and dog looking at the clouds --ar 16:9
man eating breakfast at konbini, sunlight coming in, cell shaded anime style of studio ghibli and makoto shinkai --ar 16:9Architecture
infinite gothic arch structure designed by mies van der rohe , photography by Hélène Binet --ar 16:9 --c 3
transparent recursive glass structure on the water designed by renzo piano, cinematic architectural photograph --ar 16:9
the entrance to grand central harlem station is a white, winding shape, in the style of santiago calvatrava, birds-eye-view, floating structures, extravagant settings, striped --ar 3:2Interior Design
One thing that people don't take advantage of enough: lighting conditions. As photographers and painters know, light has a huge effect on the final impression an image creates.
Describe the type of light you want to see and the time of day it is. You don't need complex vocabulary either, sometimes "large windows" is all it takes to create light pouring into a room:
Hall of a magnificent baroque palace filled with golden statues of skulls and paintings of skulls, beautiful staircase, Renaissance paintings, marble columns, high plants, large windows --ar 16:9 --s 1000
Interior by Kengo Kuma, Harmonious blend of natural elements and modern design in an eco-friendly apartment complex, falling water --ar 16:9 --c 1
Interior kitchen design by Kengo Kuma, Harmonious blend of natural elements like plants and falling water elements in an eco-friendly apartment complex --ar 16:9 --c 1Graphic Design
For practical graphic and web design applications, V6 offers a LOT more coherence in designs. I was able to get designs that were much closer to my descriptions.
Logo Design
Logos can benefit from text coherence and better understanding of the positioning of elements. Take a look at these logos:
logo for a law firm with the scales of justice above the letters "A&A", black text on white background, Rob Janoff
red and black flat logo for a kaiju head-on, flat cell shading and aggressive design, use of negative space
Marketing Illustrations
clean vector illustration of a team presenting, for a tech blog, interesting composition
very clean vector illustration of thinking about data, white background, for a tech blog, interesting composition --s 50Anything you would like to see more of? Let us know in the comments below!
Happy prompting, and I'll see you all in V7!
Hi Yubin,
great reading for the new V 6.0 Midjourney. Thank you for the effort.
Yes there is a question I struggle most of the time with is the position of the object specially with people like sitting in a curtain position, laying, jumping etc etc. Is this improved in the new V6 platform?
near that I would like to speak with you regarding your website if possible
MIMI
Very helpful, keep up the good work!