This month, Midjourney rolled out its Version 5.1.
elaborate portrait of Billie Eilish as Ghost in The Shell character by Laurey Greasley and Takeshi Obata --v 5.1
elaborate portrait of Billie Eilish as Ghost in The Shell character by Laurey Greasley and Takeshi Obata --v 5
In this study, we take an in-depth look at what the new version is capable of, what it has in common with its predecessor, and what are its superpowers and shortcomings.
MRI scan of robot samurai --v 5.1
MRI scan of robot samurai --v 5
Japanese fox god of winter death and rebirth --v 5.1
Japanese fox god of winter death and rebirth --v 5
intricate raven god --v 5.1
intricate raven god --v 5
"The only way to make sense out of change is to plunge into it, move with it, and join the dance."
— Alan Watts
Quick facts
V5.1 is (finally!) more opinionated—like V4.
It's the next step in visual aesthetics. The new model has higher coherence, better details, and improved sharpness.
Many artistic styles and techniques improved since V5.
Text prompting and Image prompting performance seemed to remain the same as in V5.
The new model ups the photorealism game (after just under two months since we saw V5!).
V5.1 generated less unwanted text. And when you want text, it's better, too.
There is an 'unopinionated' "RAW Mode" (similar to V5.0).
Finally, there is one (surprising!) issue with V5.1:
hands ¯\_(ツ)_/¯
To confirm or disprove these statements, we challenged the new model on each of these aspects.
1. Unopinionated No More
innerspection of everwrapping neutrofields of digiworld --v 5.1
innerspection of everwrapping neutrofields of digiworld --v 5
To me, this is the most exciting improvement in 5.1.
Remember, how one of the main characteristics of V5 was “unopinionatedness”? A feature that prioritized photorealism, interpreted prompts literally, and was generally more “straightforward” as opposed to “creative.“
Extreme close-up portrait --v 5.1
Extreme close-up portrait --v 5
The new model got it’s opinionatedness back, and, in that sense, combines the (improved!) intricacy of V5 with the creativeness of V4.
2. AESTHETICS + COHERENCE + DETAILS
.jpg)
Terminator movie by Satoshi Kon --v 5.1
.jpg)
Terminator movie by Satoshi Kon --v 5
What V5.1 delivers is nothing short of stunning. Overall visual aesthetics of the new model is out of this world. The images became more coherent, complex, and clear at the same time.
by John Romita Jr --v 5.1
by Angela Barrett --v 5.1
And when it comes to details, V5.1 is breathtaking. It renders highly detailed scenes more correctly, the details themselves are more pronounced, contrasted, and sharper than before.
Laurie Greasley's illustration depicting intricate mechanical heart --v 5.1
Laurie Greasley's illustration depicting intricate mechanical heart --v 5
Intricate paper cutout collage in Dada movement style --v 5.1
Intricate paper cutout collage in Dada movement style --v 5
Minas Tirith. Bas-relief --v 5.1
Minas Tirith. Bas-relief --v 5
3. ARTISTIC STYLES + TECHNIQUES
angelic apparition by Victoria Crowe --v 5.1
angelic apparition by Victoria Crowe --v 5
It’s impossible to speak of the new model’s aesthetics and not mention how it works with style modifiers (artists’ names, techniques, genres etc.) TL:DR: It’s mind-blowing!
African flora pattern by William Morris --v 5.1
African flora pattern by William Morris --v 5
Tigress Witch by Leonor Fini --v 5.1
Tigress Witch by Leonor Fini --v 5
Cute character by Go Nagai --v 5.1
Cute character by Go Nagai --v 5
There were negligibly little examples of styles that didn’t improve in one way or another compared to their predecessors (and where they didn't, we will always have V5 and --style raw ↓).
Here are just a few examples of how artistic styles and techniques in 5.1 are the next level.
In general, styles became more detailed, pronounced, and, simply put, more visually mindblowing.
by Ida Rentoul Outhwaite --v 5.1
by Ida Rentoul Outhwaite --v 5
Finally, many styles became closer to their real-life prototypes (especially noticeable with movie directors!).
by Michael Bierut --v 5.1
4. SIMPLE VS. COMPLEX PROMPTS
V5.1 ... is MUCH easier to use with short prompts.
— from Midjourney's team official announcement
15th century hero committing Leap of Faith from the wizards tower. Multi-Verse sky above, vast megacity below. Dynamic action. Extremely wide angle composition, fish-eye view, detailed illustration --v 5.1
15th century hero committing Leap of Faith from the wizards tower. Multi-Verse sky above, vast megacity below. Dynamic action. Extremely wide angle composition, fish-eye view, detailed illustration --v 5
To test how the new models reacts to simple and complex prompts and compare it to V5 in the same situation, I ran three tests, each featuring three prompts of gradually increasing complexity.
This way we can compare how prompts ranging from elementary to complex behave within both models.
Guru of Virtual Reality --v 5.1
Guru of Virtual Reality --v 5
Guru of Virtual Reality transcending time and space --v 5.1
Guru of Virtual Reality transcending time and space --v 5
Guru of Virtual Reality transcending time and space in Glitch art style --v 5.1
Guru of Virtual Reality transcending time and space in Glitch art style --v 5
stranger in wilderness --v 5.1
stranger in wilderness --v 5
stranger in wilderness of Venus --v 5.1
stranger in wilderness of Venus --v 5
stranger in wilderness of Venus on Doomsday --v 5.1
stranger in wilderness of Venus on Doomsday --v 5
mechanical trickster --v 5.1
mechanical trickster --v 5
mechanical trickster in cyberpunk world --v 5.1
mechanical trickster in cyberpunk world --v 5
mechanical trickster in cyberpunk world in Kawaii anime style --v 5.1
mechanical trickster in cyberpunk world in Kawaii anime style --v 5
How about Image prompts? To compare the two model, I fed the same set of images to both of them. And I started with my own face. 8)
in Westernpunk style --v 5.1
in Westernpunk style --v 5
Although 5.1 returns cooler results overall, the "face recognition" part didn't seem to change much. I then transitioned to try some other types of images.
.jpg)
Lady with an Ermine by Leonardo da Vinci (1489)
August Sander's photograph of famous French female scientist --v 5.1
August Sander's photograph of famous French female scientist --v 5
ultra-conceptual high-fashion neon-lit portrait of a K-Pop Diva --v 5.1
ultra-conceptual high-fashion neon-lit portrait of a K-Pop Diva --v 5
as 90's action movie poster in Akira anime style --v 5.1
as 90's action movie poster in Akira anime style --v 5
.jpg)
Amelia Mary Earhart standing in front of her plane. Chronicle (circa. 1930s)
brave female steampunk pilot in front of her plane --v 5.1
brave female steampunk pilot in front of her plane --v 5
traveling merchant of wonders and magical artefacts --v 5.1
traveling merchant of wonders and magical artefacts --v 5
Mayan priestess by Tomer Hanuka --v 5.1
Mayan priestess by Tomer Hanuka --v 5
.jpg)
Howl's Moving Castle by Hayao Miyazaki (2004)
detailed engineering infrastructural blueprint layout of Howl's Walking Castle. Technical drawing with measurements --v 5.1
detailed engineering infrastructural blueprint layout of Howl's Walking Castle. Technical drawing with measurements --v 5
bleak brutalist noir walking city by Chris Bachalo --v 5.1
bleak brutalist noir walking city by Chris Bachalo --v 5
Gino Severini's most otherworldly painting --v 5.1
Gino Severini's most otherworldly painting --v 5
This test is the ultimate illustration of how much more artistic and creative V5.1 is; how much more interesting, diverse, and detailed its images are, and how much less unwanted artefacts it generates. However, from this test I wouldn’t really give the first prize to any of the contestants. ¯\_(ツ)_/¯
Also, did you notice how V5.1 became even more photorealistic than 5?
5. PHOTOREALISM XL
Billie Eilish by Marianna Rothen --v 5.1
Billie Eilish by Marianna Rothen --v 5
Just recently, Midjourney released the V5, showcasing a revolutionary level of photorealism. It's hard to believe they could improve so much in just under two months, but they have!
19th century glass plate photograph of cybernetic ronin --v 5.1
19th century glass plate photograph of cybernetic ronin --v 5
David LaChapelle's close-up portrait of Vincent van Gogh --v 5.1
David LaChapelle's close-up portrait of Vincent van Gogh --v 5
Peter Pan. Hand-colored photograph by Lewis Hine --v 5.1
Peter Pan. Hand-colored photograph by Lewis Hine --v 5
The images in V5.1 are mind-bending, with realistic light and shadows, reflections, and skin texture.
official portrait of Star Fleet Admiral. Photograph by Nan Goldin --v 5.1
official portrait of Star Fleet Admiral. Photograph by Nan Goldin --v 5
Platon's photograph depicting Freedom Fighter --v 5.1
Platon's photograph depicting Freedom Fighter --v 5
Laure Prouvosts photograph depicting Ghandi --v 5.1
Laure Prouvosts photograph depicting Ghandi --v 5
Naturally, many styles that were photorealistic to begin with, became that much better!
by Martin Schoeller--v 5.1
by Chris Cunningham --v 5.1
by Chris Cunningham --v 5
6. ARTEFACTS + TEXT + DESIGN
1970s haute-design industrial poster with large text, flat graphics, brutalist style --v 5.1
1970s haute-design industrial poster with large text, flat graphics, brutalist style --v 5
Truly, V5.1 does dial down the amount of unwanted elements in your generations, including (mostly ;)) text, objects that "pollute" the style, and elements that depict an action instead of its result.
Pinhole photography --v 5.1
Pinhole photography --v 5
And when you do want text and symbols in your images, V5.1 usually delivers more artistic, intricate, and harmonious results. With better, more refined text lines, fonts, graphic elements (like callouts, boxes, dividers, etc.) and overall sense of design.
black-and-white brutalist infographics. 1920s graphic design by Dziga Vertov --v 5.1
black-and-white brutalist infographics. 1920s graphic design by Dziga Vertov --v 5
chart of japanese characters and chinese characters, bibliographic, 1860s letterboxing --v 5.1
chart of japanese characters and chinese characters, bibliographic, 1860s letterboxing --v 5
1970s British Punk poster with stark contrasts and playful fonts. Rebellious design Jamie Reid. Chaotic text placements, distressed letterforms, anarchic collages, iconic ransom-note typography, dissent and raw energy --v 5.1
1970s British Punk poster with stark contrasts and playful fonts. Rebellious design Jamie Reid. Chaotic text placements, distressed letterforms, anarchic collages, iconic ransom-note typography, dissent and raw energy --v 5
That said, there are quite a few cases in which V5 delivers very worthy results! If they are better or worse than 5.1 is purely a question of your goals with these prompts.
7. RAW MODE: UNOPINIONATED, AGAIN?
There is an 'unopinionated' mode for V5.1 (similar to V5 default) called "RAW Mode"
— from Midjourney's team official announcement
.jpg)
ever-adjusting neuro-enhanced artificial-intellegence-powered design --v 5.1
.jpg)
ever-adjusting neuro-enhanced artificial-intellegence-powered design --v 5
Let’s see how that works, and if RAW mode offers any advantages compared to V5, and how all that stands against the default, “opinionated” V5.1.
Instance between Birth and Death --v 5
Instance between Birth and Death --v 5.1 --style raw
Instance between Birth and Death --v 5.1
I extensively tested all three models with a set of prompts ranging from very simple to more vague and abstract ones, both leaving room for Midjourney’s imagination.
action --v 5.1 --style raw
vast green fields under alien invasion --v 5
vast green fields under alien invasion --v 5.1 --style raw
vast green fields under alien invasion --v 5.1
symphony of silent whispers in ephemeral moments weaving intricate tapestry of forgotten dreams in melancholic harmony of falling into surreal dreamscape --v 5
symphony of silent whispers in ephemeral moments weaving intricate tapestry of forgotten dreams in melancholic harmony of falling into surreal dreamscape --v 5.1 --style raw
symphony of silent whispers in ephemeral moments weaving intricate tapestry of forgotten dreams in melancholic harmony of falling into surreal dreamscape --v 5.1
As expected, nor V5, neither V5.1s RAW mode exhausted that room. ¯\_(ツ)_/¯ However, the default 5.1 showed wonders.
With marginal difference, RAW truly does inherit its unopinionated features from V5. The two models are close in visual qualities, often have same subjects, and even share some details.
portrait --v 5.1 --style raw
Some prompts didn’t really differ that much from V5 to RAW mode.
visually stunning geometrical floral pattern --v 5
visually stunning geometrical floral pattern --v 5.1 --style raw
visually stunning geometrical floral pattern --v 5.1
And there are reversed situations, where Midjourney renders more interesting results in V5 than in 5.1's RAW mode.
Goddess --v 5.1 --style raw
However, they both fade when you set them against the default 5.1. ◔__◔
8. Issues
giant's hand by Brothers Hildebrandt --v 5.1
giant's hand by Brothers Hildebrandt --v 5
V5.1 is absolutely stunning. But nothing is perfect, and there is at least one issue I need to highlight here: it’s the hands. Again.
Fortune Teller's hand by Alexandre-Evariste Fragonard --v 5.1
Fortune Teller's hand by Alexandre-Evariste Fragonard --v 5
red right hand by Rufino Tamayo --v 5.1
red right hand by Rufino Tamayo --v 5
haute-couture gloves by Guo Pei on a Princess' hands --v 5.1
haute-couture gloves by Guo Pei on a Princess' hands --v 5
But not all is lost! With a few re-rolls and some --stylize tweaking you can get very decent results. Sometimes even surpassing those from V5.
fashionable hands by Catherine Nolin --v 5.1
fashionable hands by Catherine Nolin --v 5
fashionable hands by Catherine Nolin --v 5.1 --stylize 300
fashionable hands by Catherine Nolin --v 5 --stylize 300
fashionable hands by Catherine Nolin --v 5.1 --stylize 500
fashionable hands by Catherine Nolin --v 5 --stylize 500
Conclusion
The remarkable advancements showcased in Midjourney 5.1 take the platform to new heights of aesthetic excellence, powerful default style, next-level intricacy, and broad variability.
Despite minor setbacks (bring back regular hands!), V5.1 is a huge advancement on almost every front, fixing some crucial fallbacks of its predecessor.
This extraordinary progress, at such a pace, fills me with anticipation for the future of Midjourney. And it's around the corner, take it from the Midjourney team themselves:
There may be further tunings of V5.1 styles
and possibly a V5.2 after that
— from Midjourney's team official announcement
Happy midjourneys,
— Andrei Kovalev
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! →
/explore Midjourney styles
.jpg)
.jpg)
.jpg)
.jpg)
.jpg)
.jpg)
.jpg)
.jpg)
.jpg)
.jpg)
