/blend looks at the 'concepts' and 'vibes' of your images and merges them together into novel interpretations.

— from Midjourney's official Discord

Last week, Midjourney made the official announcement about adding a new feature to their Discord bot that allows blending two to five images together via a handy interface where you upload images directly instead of pasting their public URLs.

In this study, we will take an in-depth look at how /blend works and will try to take it to its limits!

MULTI-IMAGE PROMPT, BUT EASIER

The blending feature has existed as Multi-Image Prompt since V4 came out, and we even tested it already in Quick Look At Midjourney V4. However, when trying the /blend interface, I noticed that it became more advanced and capable.

I used either my own photographs, my Midjourney-generations, or classical works of art for this study.

[IMAGE 1]
Francis D.

[IMAGE 2]
painting by Asger Jorn depicting 19th century glowing dieselpunk space explorer --ar 2:3

[BLEND]

[IMAGE 2]
intricate cat mask by Anton Corbijn. Layout. Scientific photograph. Marks and notes --test --creative

[BLEND]

[IMAGE 2]
catholic icon mecha --test --creative

[BLEND]

[IMAGE 2]
Vladimir Svirsky for Pyotr Fomenko Workshop Theatre

[BLEND]

[IMAGE 2]
Nick Cave by Maria Lassnig

[BLEND]

Whereas before, Midjourney was "afraid" of touching/altering significant facial features from an initial photograph (eyes, nose, mouth, and skin in general), it is now much more "daring!"

[IMAGE 2]
A masks for a personal project

[BLEND]

In short, Multi-Image Prompting is an almost all-powerful tool for blending any images together. And /blend just made its power more accessible. No more uploading images and copying and pasting URLS! You simply drag the pictures into their slots and press "play."

You can also explicitly set the aspect ratio. Otherwise, your blends will be square. Note that they will often inherit the framing from one or both initial images. So if you're using a 2:3 picture and don't want black stripes on your generation sides, use the Dimension parameter to choose the blend's aspect ratio.

Let's see how it works with some real-world examples. (Spoiler: it's impressive!)

CHARACTER CREATION

I talked about this in Midjourney For Photographers—you can use Image Prompts to see how a particular person might look in your future photoshoot.

Imagine, you are a casting director for a movie. You send a request to a casting agency. They return photographs of candidates. Now, before showing them to the director, you use Midjourney's /blend to generate stills from a movie—long ahead of the actual shooting. Let's try it with Francis D. and a casting director working on a Viking movie.

First, you use a simple Image Prompt to generate a basic character. It's good, but then you may want to take him up a notch and expand your prompts with cinematic details.

[IMAGE PROMPT]
(Francis D.) as Viking with entangled beard

[IMAGE PROMPT]
(Francis D) as Viking with beard icy lake in background. Wide shot by Roger Deakins

I used the latter variant for the following tests.

Let's add more context and epicness, but this time use /blend to mix our character with a cinematic MJ generation from the Vikings' life. ]:-)

[IMAGE 2]
viking standing on drakkar on moonlit night. Closeup shot by Roger Deakins

[BLEND]

The director sent you as a reference—a still from the Vikings TV show. You bring the director's vision closer to life—with some /blend magic. And Francis D.!

[IMAGE 2]
Jarl Borg played by Georgia Hirst. Vikings TV show (2013–2020)

[BLEND]

Finally, you ignore the /blend interface and use Image Prompt for blending two images with additional text part.

[IMAGE 2]
Jarl Borg played by Georgia Hirst. Vikings TV show (2013–2020)

[BLEND]

You just made the director a happy person for one instance!

Of course, he would still want to see the candidates in action. But proofs-of-concept—i.e., Midjourney blends—can considerably help with the initial selection.

THE CUTE MAINECOON CAT

Let's try /blend on something utterly different from human faces. Meet Maddy (short for Mad Love ^___^)—the inspiration behind the cute Mainecoon cat test from our benchmark.

[IMAGE 1]
Maddy ^_____^

[IMAGE 2]
black Siberian cat-samurai with baroque antlers. Closeup portrait. 1800s highly detailed encyclopedia etching --test --creative

[BLEND]

I tried combining Maddy's portrait with three MJ generations of different styles and complexity. Note how easily strong stand-out features transfer to the final generations (the horns, the shapes and paint from Optimus Prime, the saturated red spot on the fashionable cat's nose).

[IMAGE 2]
Optimus Prime cat --test --creative

[BLEND]

[IMAGE 2]
cute Mainecoon cat by Erik Madigan Heck

[BLEND]

As you can see, /blend works great with animals. But how about non-portraits?

OBJECTS, PLACES, ABSTRACTS

First, I tried blending some of my real-life masks with the Midjourney-generated ones.

[IMAGE 1]
A Berikaoba ritual mask for a personal project

[IMAGE 2]
dieselpunk samurai beetle mask headpiece --test --creative

[BLEND]

[IMAGE 1]
A Post-apocalyptic Orca Samurai mask for a personal project

[IMAGE 2]
anime girl wearing intricate fish mask by Tim Walker

[BLEND]

[IMAGE 1]
A sheep skull mask for a personal project. 3D render from a real-life object's photogrammetry

[IMAGE 2]
full page scan of an antique lithograph of fantasy mammoth skull on black paper. Baroque motives, ornate, intricate design --test --creative

[BLEND]

As a mask designer, I strongly encourage fellow artists from that domain of art to try Midjourney's /blend with their works. It is an extraordinary tool for finding inspiration and generating ideas for your next pieces based on existing ones.

Okay, we've seen that /blend works great with portraits and masks. What about still life? It works excellently.

[IMAGE 1]
A personal study in photographing still-life

[IMAGE 2]
Jan Brueghel the Elder — Bouquet (1599)

[BLEND]

[IMAGE 1]
Abandoned film development lab at Georgian Film Studios (Tbilisi)

[IMAGE 2]
Scene from "The Shining" (1980). Cinematographers: John Alcott, Martin Kenzie

[BLEND]

And interiors? Yes.

Finally, let's use /blend to create abstract art! I used a photograph of a vintage film tripod detail shot for a personal project as one entry and Vladimir Tatlin's drawing of Tatlin's Tower—a masterpiece of Constructivist art—as another.

[IMAGE 1]
A vintage movie tripod detail for a personal project

[IMAGE 2]
Vladimir Tatlin, Tatlin's Tower (1919)

[BLEND]

UP TO FIVE IMAGES, SERIOUSLY?

For the post-scriptum of this chapter, I wanted to see how adding more than one image into the mix works. I used three of my MJ works for the first blend, and was adding one new image to each of the following ones.

[IMAGE 1]
deadly atompunk female Archbishop. Hand-colored photographic print --test --creative

[IMAGE 2]
floral beauty by Sarah Moon --test --creative

[IMAGE 3]
korean folk theatre fish mask --stylize 3750 --test --creative

[BLEND]

[IMAGE 4]
dark Japanese industrial heavy graphic poster with mecha bird by Anton Corbijn --test --creative

[BLEND]

[IMAGE 5]
full page scan of highly detailed 1800s drawing of cyberpunk monk. Desaturated hues --test --creative

[FINAL BLEND FROM 5 IMAGES]

The results are inspiring to say the least! /blend manages to keep significant features from all five images—and does it consistently as you go from two to five!

IMPORTANT: Naturally, I tried adding several instances of the same image into /blend to control the "weight" of the image in the mix. I learned that adding the same image more than once is restricted in Midjourney, and trying to bypass it can end in revoking your access!
(●＞ω＜●)

"TEACH" MIDJOUNREY A NEW ARTISTIC STYLE

It often happens that Midjourney doesn't know a specific artist or technique. I used /blend to bring a new style into Midjourney and rework an existing image in that style. And I started with a painter whose name is unknown to Midjourney (sadly)—Adrian Allinson.

[IMAGE 1]
red rabbit

[IMAGE 2]
Adrian Allinson — Spring In Elba

[BLEND]

MJ still struggles quite often to rework a photorealistic image into an illustration or drawing.

[IMAGE 1]
Legendary translator Victor Golyshev for a personal project

[IMAGE 2]
Vincent Van Gogh — The Zouave (1888)

[BLEND]

However, there is a trick! I used Fast Style Transfer AI model in Phygital+ to roughly translate the drawing style to the initial photo.

Phygital+ interface with nodes connecting a style reference with target image to a Fast Style Transfer model

Then I fed the result to Midjourney via /blend, and it worked much better than with the photograph.

[IMAGE 1]
Fast Style Transfer output image

[IMAGE 2]
Vincent Van Gogh — The Zouave (1888)

[BLEND]

/BLEND VS. TEXT PROMPTS

Finally, a lot of the strongest and most beautiful Midjourney styles are not exactly the same as the original styles. Let's use /blend alongside traditional prompts to see how both work to bring a style into our generations.

[ORIGINAL IMAGE]
Still frame from my video for Jack The Ripper-themed audio play promotion

[IMAGE PROMPT]
by László Moholy-Nagy

[BLEND IMAGE 2]
László Moholy-Nagy — Large Railway Painting (1920)

[BLEND RESULT]

On the other hand, some artists' styles are too diverse, and Midjourney might only know one part of their body of works. Take German painter Jeanne Mammen, for example. Her art developed and changed drastically throughout her artistic career. Early in her career, Mammen's work was heavily influenced by Art Nouveau. Her magazine and book illustrations of the time often featured ornate, decorative elements and women in elegant, flowing clothing.

by Jeanne Mammen --v 4

As her career progressed, Mammen started experimenting with new techniques and materials, and her work became more expressive and dynamic, shifting to New Objectivity. Her paintings from the 1920s and 1930s featured bold, abstract shapes and focused on movement, and charged with social and political commentary.

Midjourney mostly knows the very first Mammen, generating light painterly illustrations in soft, subdued hues with occasional splashes of color. Here is how Jeanne Mammen's name works with Image Prompt and what you can do with /blend using original artwork from her early career.

[IMAGE 1]
Francis D. makes a comeback ;)

[IMAGE PROMPT]
(Francis D.) by Jeanne Mammen

[IMAGE 2]
Jeanne Mammen — She repre­sents (ca. 1928)

[BLEND]

But what about her later—and completely different—works that Midjourney doesn't seem to know? Here is what happens if you blend Francis D. with a fragment of Mammen's powerful abstract painting from the early 1940s.

[IMAGE 2]
Jeanne Mammen, Dying Warrior (Young Soldier in the Front Fire), ca. 1943

[BLEND]

There is another example of how a "style inception" works in MJ. Let's take a wildly talented yet unknown to Mirdjourney photographer—me. 0: ) In this study, I used many of my own photographs. And Midjourney brought at least some part of my style into each final image.

As an artist, I feel grateful for such a mind-altering creative opportunity.

Happy midjourneys!

— Andrei