Midjourney's Style Reference (--sref) for Photographers | In-depth Guide | Andrei Kovalev's Midlibrary (2024)

So, in this guide, we will explore how the Style Reference works within the context of photography and how photographers can use it to enhance their artistic practice. And we will do so through the lens of my own pictures.

Let’s get to it!

How --sref works

Utilizing the Style Reference is quite straightforward: simply add --sref followed by the URL of a publicly accessible image after the text part of your prompt. If you wish to reference multiple images, add their URLs consecutively, separated by aspace.

An important parameter to consider is the Style Weight, or --sw. It determines how much the referenced style will influence the outcome, and ranges from 0 to 1000, with the default setting being 100.

To put these features to the test, let’s use a cinematic photograph of the artist Oleg Paschenko as our Style Reference, maintain a consistent seed value, and then progressively adjust the --sw parameter from 1 to 1000 to observe its effect on the results.

For this experiment, I also mentioned an artistic style in the prompt itself. This approach allows one end of the spectrum to be tied to a renowned movie director’s style, while the other end showcases the maximal influence of the reference image.

Now that we're done with technicalities, let's add some more photographic styles into the mix!

Referencing Photographic Styles

Midjourney’s Style Reference draws upon the visual elements, or character, of the original image, such as colors, composition, texture, background, and even the overall atmosphere.

Details like costumes, outfit elements, props, and accessories often merge into the resulting images.

Occasionally, --sref might also incorporate the original subject (or their stand-out features) into the output. Unless the subject is specifically defined in the prompt.

Through experimenting with my portraits, I've observed that close-ups bring better results than photographs featuring more complex scenes, set-designs, or costumes.

However, persistence often helps unveil the strong sides of the reference style—rembember, certain prompts might be more receptive to specific references than others.

Another finding is that less “pronounced,” softer photographs with heavy visual effects (e.g., shallow depth of field, blurry segments, lens flares, etc.) tend to produce more favorable outcomes.

How about the use of more complex, semi-photographic styles? To answer this, I conducted another experiment. I selected a scan of a collage I had created years ago, which incorporates a photograph of my friend and muse, the Georgian theatre and cinema actor Malkhaz Abuladze. To take this test even further, Ifirst tried to add the original photo into the blend using Image Prompt.

Initially, I utilized the original portrait as an Image Prompt and the collage as the Style Reference. Here are three outputs, each resulting from different parameter settings: the first one showcases the default result. In the second, I increased the Image Weight (--iw) to its maximum to align the outcome as closely with the original as possible. For the third scenario, I maxed out the Style Weight to its highest, granting full freedom to the impact of --sref.

An important aspect to keep in mind is that you can apply a Style Reference showcasing a specific subject or shot in one genre to prompts that significantly diverge from the original’s theme.

To demonstrate this, we'll use a still-life photograph as the Style Reference to see how it adapts across a range of prompts: from another still-life, which greatly differs from the initial one thematically, to documentary and fashion photography.

And did Imention that you can use more than one image as Style Reference?

Multi-Style References

Let’s now try to use more than one style reference—and pick contrasting images for this experiment.

The sheer number of variants—creative possibilities—unlocked by this mechanism is… staggering. Essentially, blending any combination of contrasting images can forge a unique (and often surprising!) style by drawing powerful elements from both inputs.

And this is what happens if we use three reference images instead of two:

While employing multi-reference styles with more than two contrasting references can “dilute” the distinct characters of the original images, it also allows for wider range of promptsto successfuly reproduce prominent features of the source photographs.

Now, we've experimented with Style References that are specific photographs depicting particular subjects. But here's a little trick to elevate your neurophotography game—this time, referring to abstract, non-specific, “technical” sources.

Referencing ”Technical” Styles

If you've ever attempted to incorporate “rough” photographic effects—like film grain, dust and scratches, light leaks, and so on—into your prompts, you might have noticed that Midjourney isn’t very good at being “low-quality.”

The images with these effects tend to be too refined and polished to resemble their real-life counterparts accurately. Now, --sref is here to solve this problem!

A Style Reference can be just a color combination—like this circular gradient:

And don't forget we can mix those references together. So what will the grain, the light leak, and the pshychedelic gradient bring if referred together at the same time?

Similarly, we can use other—more “pronounced“ analogue effects references, like burned film texture...

...or, for instance, a VHS-glitch:

But, we've almost stirred away from the subject of this guide—Style References in Photography.

Referencing Existing Styles

The Style Reference is great for summoning styles that aren’t known to Midjourney or are too complex to easily achieve with a text prompt. Here are three examples showcasing such powerful, yet previously elusive, styles.

The Prokudin-Gorskii Case

Sergey Prokudin-Gorskii (1863–1944) was an early pioneer of color photography who produced a vast body of work at the turn of the 20th century. A chemist, an engineer, and a devoted photographer, Prokudin-Gorskii traveled across the Russian Empire—sometimes by a special train, sometimes by a steamboat—to document its life and culture. His work includes several hundred color photographs, each a unique document of its time.

Original images: courtesy of Library Of Congress.

Using an innovative technical process, Prokudin-Gorskii produced color photographs long before color film was invented. As one of the earliest precursors to modern-day RGB color processing, he would take three separate black and white photos of the same scene, using a separate color filter — red, green, and blue — for each.

When projected through their respective filters in the darkroom, these negatives together produced a color image!

As photographer, I have been inspired by Prokudin-Gorskii's process for a long time and have used it in my own projects numerous times. However, to my disappointment, when I tried Prokudin-Gorskii's name in Midjourney, I found that my favorite AI doesn't recognize the style of my favorite photographer.

But now, with --sref, can we use it to bring the maestro’s visual style to Midjourney? Let’s try using three of Porkudin-Gorskii’s most famous works.

And the result doesn't disappoint...

The Criminal Case

For this part, we will travel into the world of vintage mugshots. These arresting images, captured by New South Wales Police Department photographers in the 1920s and 1930s, not only reveal a slice of criminal history but also provide us with a beautiful piece of art, although probably not created with art in mind.

Original images: courtesy of Australia’s Justice & Police Museum.

Let’s see if these portraits, released by the Australia’s Justice & Police Museum↗︎ can help us summont that unique visual style in our prompts.

And once again, some truly stunning results!

The Classical Case

Original images: courtesy of Studio Harcourt.

Chances are, you’ve seen the photographs of the legendary Studio Harcourt many times before. Throughout the 20th century, Harcourt’s photographers produced hundreds of iconic portraits of some of the most famous men and women around the world.

And the style of these photograph became one of the most recognizable portrait styles in art history. Well, not to Midjourney, apparently ¯\_(ツ)_/¯ Yes, for the “by Studio Harcourt” prompt, our beloved MJ returns a ”generic“— a set of nice looking pictures having nothing to do with the renowned Studio.

Well, let’s try to refer to some of Harcourt’s iconic works to bring their style into Midjourney.

And we have a 100% hit yet again.

Style References +Image Prompts

For the concluding part of our guide, let’s dive into the alchemy of using existing images for both, prompts and Style References.

The purpose here is straightforward: to transplant a style from our reference onto another image. For this series of tests, Iwill come back to my own portraits as sources for Image Prompts, but will use three different kinds of style references.

The first pair is a portrait of my friend and muse Francis D. (a regular in our guides ;)), and a random composition I came up with after five minutes in Photoshop.

The second pair is a portrait I made for a personal project in Tbilisi, combined with an editorial portrait of a renowned winemaker (as Style Reference).

And for the final trial,Ipicked a writer's press portrait and combined it with aniconic historical portrait of a legendary Arctic explorer.

Style References for Photographers

--sref is an exceptionally powerful feature, allowing us to extract visual styles from existing images to apply them to our Midjourney prompts. But how can we, as photographers, leverage this feature in our artistic practice?

With Midjourney’s Style Reference, we can generate unique photo-shoot concepts, remix inspirations into novel ideas and research unexpected visual styles, previsualize projects with unprecedented detail, generate mood boards, set design references, and styling options, reimagine archival photographs, and ultimately push the boundaries of traditional photography towards a new exciting genre—neurophotography.

Happy midjourneys,
—Andrei