Prompt Modifiers (Sample Placeholder Data)

Table of Contents

Midjourney has released a new feature --sref that allows you to pull a style from one image and apply it to another. Let’s dive in!

Welcome to the latest buzz in the AI art world: Midjourney’s groundbreaking “Style Reference” feature, a true game-changer for creators seeking consistency in their digital masterpieces. On January 31, 2024, David Holz, the visionary CEO of Midjourney, unveiled this eagerly anticipated functionality, igniting waves of excitement in the Midjourney universe.

Imagine being able to infuse your artwork with the distinctive style of any image that inspires you by simply including its URL in your prompt. That’s the magic of the new Style Reference feature. It intelligently extracts the essence of style — and even some color nuances — from your chosen reference images and seamlessly blends them into your creations. This isn’t about vague influences or hit-and-miss attempts to capture a certain aesthetic. It’s about precision, directly harnessing the soul of your favorite styles and making them an integral part of your art.

Now, I’ve been down the rabbit hole of trying to achieve consistent styles in my AI-generated art, and let’s say it’s been a quest filled with trials and errors. Until now, that is. Introducing Style References marks a significant leap forward from the previous style tuners we’ve tinkered with. Don’t get me wrong, the older style tuners have their charm (and you can dive into my past articles to explore that saga), but they pale in comparison when it comes to the sheer capability of Style References to maintain a stylistic throughline across varied prompts.

It’s worth noting that this shiny new tool is exclusive to version 6 of Midjourney, setting it apart from last year’s Style Tuner, which was the hallmark of version 5.2. But don’t let that stop you from getting creative — there’s nothing to prevent us from crafting our own reference images to serve as style guides. The difference is night and day, with Style References offering a level of stylistic fidelity that truly transforms the art creation process.

How to use Style References (From the Midjourney announcement)

Type — sref after your prompt and put one (or more) URLs to images like this — sref urlA urlB urlC

The image model will look at the image urls as ‘style references’ and try to make something that ‘matches’ their aesthetics

Set relative weights of styles like this — sref urlA::2 urlB::3 urlC::5

Set the total strength of the stylization via — sw 100 (100 is default, 0 is off, 1000 is maximum)

Regular image prompts must go before — sref like this/imagine cat ninja ImagePrompt1 ImagePrompt2 — sref stylePrompt1 stylePrompt2

This works for both V6 and Niji V6 (it does not work with V5 etc)

(please note that the long dashes above in the commands are actually: -- Medium likes to “fix” them for us. )

Getting Started

First step. Use the /settings command in the chat bar where you generate or with the Midjourney bot directly. Make sure that you have version 6 selected. It currently says [ALPHA] next to it, but that will likely change soon, but as long as it is v6, it is fine.

Next, you need a URL for an image. If you do not have a web link to an image, you will want to upload your image to Discord and get a URL from there.

Uploading an Image to Discord for Midjourney

  1. Send your image to the Midjourney Bot in Discord. Paste the image into the chat and hit Enter, or use the + symbol at the left of the text bar and navigate to a saved file on your local file system.
  2. To get the direct image link:
  • Desktop App: Right-click the image and choose “Copy Link.”
  • Web App: Click to expand the image, right-click, and select “Copy image address.”
  • Mobile App: Press and hold the image, then tap “Copy Media Link.”
  • Alternatively, click to expand the image in Discord and choose “Open in Browser” at the bottom, then copy the URL directly from your browser.

One reference image at a time

Let’s begin with something distinctive to see what influence an image has on our prompt.

URL: https://cdn.midjourney.com/u/b1010388-32a5-4633-abc3-57962c9d86de/1a9de9dc053380449f6ae932b72eb62827d39bf8cc188d84f57b12c9af179bc1.png

<span id="7624" data-selectable-paragraph="">a basket full of playful kittens --sref https://s.mj.run/ln_ahmdE43c --v 6.0</span>

While I did pop in the full URL for that earlier image, Midjourney went ahead and trimmed it down in their response. But no worries, it still points to the same picture. If you give that link a click, it’ll stretch right back out to the full URL in your browser.

I took the same prompt and played around with the style weight — sw, cranking it up to 200, then 500, and all the way to 999. This style weight thing is pretty neat; it ramps up the influence of your style reference on your prompt. I’m still getting the hang of it, so I can’t dish out any solid advice just yet. From what I’ve seen, going too high can sometimes make the image a bit less coherent, but I reckon that’s just part of the learning curve with new tools like this. Plus, I’m expecting the results to get more predictable as Midjourney fine-tunes this feature. And, of course, how it all turns out really depends on the mix of your style reference and prompt.

<span id="0eae" data-selectable-paragraph=""><span>a</span> basket full of playful kittens <span>--sref</span> https://s.mj.run/frTdH7J_Gio --v <span>6.0</span></span>

I took the same prompt but spiced it up with a style weight — sw, setting it at 200 on the left, 500 in the middle, and going all out with 999 on the right. The changes weren’t as dramatic as I thought they’d be this time. Could be just the mix of prompt and style reference I picked.

<span id="d233" data-selectable-paragraph="">a basket full of playful kittens --sref https://s.mj.run/O_w9neuqCMI --v 6.0</span>

Again below with the same prompt as above but adding a style weight --sw of 200 (left), 500 (middle), and 999 (right).

<span id="55ab" data-selectable-paragraph=""><span>a</span> <span>basket</span> <span>full</span> <span>of</span> <span>playful</span> <span>kittens</span> <span>--sref</span> <span>https</span>:<br><span>a</span> <span>basket</span> <span>full</span> <span>of</span> <span>playful</span> <span>kittens</span> <span>--sref</span> <span>https</span>:<br><span>a</span> <span>basket</span> <span>full</span> <span>of</span> <span>playful</span> <span>kittens</span> <span>--sref</span> <span>https</span>:</span>

Getting fancy and combining

I prepared those three examples earlier to set the stage for this mix-and-match experiment. Our first example was all about pointillism, dotted to perfection. The second had that cosmic, swirling vibe, while the third was a dive into an organic, floral world.

I took our base prompt for another spin three times to check out the natural play of variations when we blend those style references. The image on the right? It’s brimming with that cosmic feel. Yet, all three are speckled with vibrant pops from our dotty first example. And as predicted, the elements from examples one and three — the dots, leaves, and butterflies — really came together, each getting its moment in the spotlight. This trio is all about equal opportunity for each style, and the result? It is a fascinating collage of spots, nature, and cosmic swirls.

<span id="fc77" data-selectable-paragraph="">a basket full of playful kittens --sref https://s.mj.run/ln_ahmdE43c https://s.mj.run/Lw_0fjfp0TM https://s.mj.run/O_w9neuqCMI --v 6.0</span>

This time around, I played with the weightings of our trio of style references, tipping the scales in favor of our spacey example two — giving it twice the influence of the others. Despite amping up the cosmic vibes, we still catch glimpses of the other elements. However, the space-themed curves and wavy lines are now taking center stage, making a bolder statement in the mix.

a basket full of playful kittens — sref https://s.mj.run/ln_ahmdE43c ::3 https://s.mj.run/Lw_0fjfp0TM ::6 https://s.mj.run/O_w9neuqCMI ::3 — v 6.0

a basket full of playful kittens — sref https://s.mj.run/ln_ahmdE43c ::3 https://s.mj.run/Lw_0fjfp0TM ::6 https://s.mj.run/O_w9neuqCMI ::3 — v 6.0

<span id="e28a" data-selectable-paragraph="">a basket full of playful kittens --sref https://s.mj.run/ln_ahmdE43c ::3 https://s.mj.run/Lw_0fjfp0TM ::6 https://s.mj.run/O_w9neuqCMI ::3 --v 6.0 </span>

Below, I doubled the value of example three over the other two. Notice we lose the spacey feel and have many more organic plant elements now.

https://s.mj.run/ln_ahmdE43c ::3 https://s.mj.run/Lw_0fjfp0TM ::3 https://s.mj.run/O_w9neuqCMI ::6 — v 6.0

https://s.mj.run/ln_ahmdE43c ::3 https://s.mj.run/Lw_0fjfp0TM ::3 https://s.mj.run/O_w9neuqCMI ::6 — v 6.0

<span id="a41e" data-selectable-paragraph="">a basket full of playful kittens --sref https://s.mj.run/ln_ahmdE43c ::3 https://s.mj.run/Lw_0fjfp0TM ::3 https://s.mj.run/O_w9neuqCMI ::6 --v 6.0</span>

I took a gamble and experimented with a negative weight, not really expecting it to take effect, but to my surprise, it did! In the image below, I assigned a -3 weight to our spacey reference while bumping up the influence of our third example image, and the transformation was striking. It’s a revelation that opens up a whole new dimension of possibilities. I’m intrigued by what this could mean and it’s definitely something I’ll be exploring in more depth in a future piece — there’s a whole world of potential just waiting to be uncovered here.

https://s.mj.run/ln_ahmdE43c ::3 https://s.mj.run/Lw_0fjfp0TM ::-3 https://s.mj.run/O_w9neuqCMI ::6 — v 6.0

a basket full of playful kittens — sref https://s.mj.run/ln_ahmdE43c ::3 https://s.mj.run/Lw_0fjfp0TM ::-3 https://s.mj.run/O_w9neuqCMI ::6 — v 6.0

<span id="b804" data-selectable-paragraph="">a basket full of playful kittens --sref https://s.mj.run/ln_ahmdE43c ::3 https://s.mj.run/Lw_0fjfp0TM ::-3 https://s.mj.run/O_w9neuqCMI ::6 --v 6.0 </span>

Now let’s give the example one reference image extra weight:

a basket full of playful kittens — sref https://s.mj.run/ln_ahmdE43c ::6 https://s.mj.run/Lw_0fjfp0TM ::3 https://s.mj.run/O_w9neuqCMI ::3 — v 6.0

<span id="ea44" data-selectable-paragraph="">a basket full of playful kittens --sref https://s.mj.run/ln_ahmdE43c ::6 https://s.mj.run/Lw_0fjfp0TM ::3 https://s.mj.run/O_w9neuqCMI ::3 --v 6.0</span>

We see some quirks in these images, like the peculiar eyes in the second one. It’s important to remember that these fluctuations are part of the ongoing improvements, especially since Midjourney v6 is still in its alpha phase, meaning it’s updated almost daily. Given these rapid changes, I’d suggest holding off on using this for any crucial design projects that require consistency. For now, v5.2 might be a safer bet. What you prompt today in v6 alpha might evolve significantly by next week, so staying flexible is key.

Even more ways to combine!

We can also mix in image prompts with our style references! I brought Kitty Miranda from a previous piece into this creative blend.

Kitty Miranda — created for a previous article

To test out the final part of the announcement, I will add an image prompt of Kitty Miranda to our weighted style image reference prompts we were using above.

a basket full of playful kittens https://s.mj.run/43bAPM-ATlw — sref https://s.mj.run/ln_ahmdE43c ::6 https://s.mj.run/Lw_0fjfp0TM ::3 https://s.mj.run/O_w9neuqCMI ::3 — v 6.0

<span id="508a" data-selectable-paragraph="">a basket full of playful kittens https://s.mj.run/43bAPM-ATlw --sref https://s.mj.run/ln_ahmdE43c ::6 https://s.mj.run/Lw_0fjfp0TM ::3 https://s.mj.run/O_w9neuqCMI ::3 --v 6.0 </span>

Notice how Kitty Miranda has lowered the number of playful kittens in our basket. It didn’t give us her fun fruit hat, although #3, bottom left, appears to have been trying.

a basket full of playful kittens https://s.mj.run/43bAPM-ATlw ::6 — sref https://s.mj.run/ln_ahmdE43c ::6 https://s.mj.run/Lw_0fjfp0TM ::3 https://s.mj.run/O_w9neuqCMI ::3 — v 6.0

a basket full of playful kittens https://s.mj.run/43bAPM-ATlw ::6 — sref https://s.mj.run/ln_ahmdE43c ::6 https://s.mj.run/Lw_0fjfp0TM ::3 https://s.mj.run/O_w9neuqCMI ::3 — v 6.0

<span id="dc9a" data-selectable-paragraph="">a basket full of playful kittens https://s.mj.run/43bAPM-ATlw ::6 --sref https://s.mj.run/ln_ahmdE43c ::6 https://s.mj.run/Lw_0fjfp0TM ::3 https://s.mj.run/O_w9neuqCMI ::3 --v 6.0 </span>

It is possible to use the images and style reference images without weights, or with the image prompt weighted and the rest not. The variations are limitless.

a basket full of playful kittens https://s.mj.run/43bAPM-ATlw ::10 — sref https://s.mj.run/ln_ahmdE43c https://s.mj.run/Lw_0fjfp0TM https://s.mj.run/O_w9neuqCMI — v 6.0

a basket full of playful kittens https://s.mj.run/43bAPM-ATlw ::10 — sref https://s.mj.run/ln_ahmdE43c https://s.mj.run/Lw_0fjfp0TM https://s.mj.run/O_w9neuqCMI — v 6.0

<span id="7769" data-selectable-paragraph="">a basket full of playful kittens https://s.mj.run/43bAPM-ATlw ::10 --sref https://s.mj.run/ln_ahmdE43c https://s.mj.run/Lw_0fjfp0TM https://s.mj.run/O_w9neuqCMI --v 6.0</span>

I decided to try a slightly different prompt for Kitty Miranda using varied weights on our style reference images:

A cat in a large hat made of tropical fruit. https://s.mj.run/43bAPM-ATlw — sref https://s.mj.run/ln_ahmdE43c ::3 https://s.mj.run/Lw_0fjfp0TM ::2 https://s.mj.run/O_w9neuqCMI ::6 — v 6.0

A cat in a large hat made of tropical fruit. https://s.mj.run/43bAPM-ATlw — sref https://s.mj.run/ln_ahmdE43c ::3 https://s.mj.run/Lw_0fjfp0TM ::2 https://s.mj.run/O_w9neuqCMI ::6 — v 6.0

<span id="b0fb" data-selectable-paragraph="">A cat in a large hat made of tropical fruit. https://s.mj.run/43bAPM-ATlw --sref https://s.mj.run/ln_ahmdE43c ::3 https://s.mj.run/Lw_0fjfp0TM ::2 https://s.mj.run/O_w9neuqCMI ::6 --v 6.0</span>

I suspect that my three different combined style reference images are simply too diverse, given the results.

A cat in a large hat made of tropical fruit. https://s.mj.run/43bAPM-ATlw ::2 — sref https://s.mj.run/ln_ahmdE43c ::6 https://s.mj.run/Lw_0fjfp0TM ::2 https://s.mj.run/O_w9neuqCMI ::10 — v 6.0

A cat in a large hat made of tropical fruit. https://s.mj.run/43bAPM-ATlw ::2 — sref https://s.mj.run/ln_ahmdE43c ::6 https://s.mj.run/Lw_0fjfp0TM ::2 https://s.mj.run/O_w9neuqCMI ::10 — v 6.0

<span id="fafc" data-selectable-paragraph="">A cat in a large hat made of tropical fruit. https://s.mj.run/43bAPM-ATlw ::2 --sref https://s.mj.run/ln_ahmdE43c ::6 https://s.mj.run/Lw_0fjfp0TM ::2 https://s.mj.run/O_w9neuqCMI ::10 --v 6.0</span>

To avoid mixing such diverse styles, I have paired Kitty Miranda with just the prompt, the image prompt with a weight of 2, and the style reference image #2 with the abstract space image.

A cat in a large hat made of tropical fruit. https://s.mj.run/43bAPM-ATlw ::2 — sref https://s.mj.run/Lw_0fjfp0TM — v 6.0

A cat in a large hat made of tropical fruit. https://s.mj.run/43bAPM-ATlw ::2 — sref https://s.mj.run/Lw_0fjfp0TM — v 6.0

<span id="8782" data-selectable-paragraph="">A <span>cat</span> <span>in</span> a large hat made of tropical fruit. https://s.mj.run/43bAPM-ATlw ::2 --sref https://s.mj.run/Lw_0fjfp0TM --v 6.0 </span>

Upon reaching this step, I remembered the style weight command and decided to give it a try. I was not disappointed as the spacey swirls became more prominent.

A cat in a large hat made of tropical fruit. https://s.mj.run/43bAPM-ATlw — sref https://s.mj.run/Lw_0fjfp0TM — sw 777 — v 6.0

<span id="c1a5" data-selectable-paragraph="">A cat in a large hat made of tropical fruit. https://s.mj.run/43bAPM-ATlw --sref https://s.mj.run/Lw_0fjfp0TM --sw 777 --v 6.0 </span>

I decided it was time to revisit the other two style reference images for better coverage. Here, I have replaced the sref from the prompt above with the example one image that showcases pointillism.

<span id="9c8a" data-selectable-paragraph="">A cat in a large hat made of tropical fruit. https://s.mj.run/43bAPM-ATlw --sref https://s.mj.run/ln_ahmdE43c --sw 777 --v 6.0  </span>

I have returned to example three, which has a natural leafy style.

A cat in a large hat made of tropical fruit. https://s.mj.run/43bAPM-ATlw — sref https://s.mj.run/O_w9neuqCMI — sw 777 — v 6.0

A cat in a large hat made of tropical fruit. https://s.mj.run/43bAPM-ATlw — sref https://s.mj.run/O_w9neuqCMI — sw 777 — v 6.0

<span id="1411" data-selectable-paragraph="">A cat in a large hat made of tropical fruit. <a href="https://s.mj.run/43bAPM-ATlw" rel="noopener ugc nofollow" target="_blank">https://s.mj.run/43bAPM-ATlw</a> --sref <a href="https://s.mj.run/O_w9neuqCMI" rel="noopener ugc nofollow" target="_blank">https://s.mj.run/O_w9neuqCMI</a> --sw 777 --v 6.0</span>

I noticed the style didn’t transfer well when using an image reference. Therefore, I removed the reference and obtained a more natural-looking style.

A cat in a large hat made of tropical fruit. — sref https://s.mj.run/O_w9neuqCMI — sw 999 — v 6.0

A cat in a large hat made of tropical fruit. — sref https://s.mj.run/O_w9neuqCMI — sw 999 — v 6.0

<span id="b1ff" data-selectable-paragraph="">A cat in a large hat made of tropical fruit. --sref https://s.mj.run/O_w9neuqCMI --sw 999 --v 6.0</span>

Since I removed Kitty Miranda from the prompt above, I decided to see what would happen if I turned her into part of the sref as a style reference image. It still gives that natural leafy feel but not as closely as I might have hoped for.

A cat in a large hat made of tropical fruit. — sref https://s.mj.run/O_w9neuqCMI https://s.mj.run/43bAPM-ATlw — sw 999 — v 6.0

A cat in a large hat made of tropical fruit. — sref https://s.mj.run/O_w9neuqCMI https://s.mj.run/43bAPM-ATlw — sw 999 — v 6.0

<span id="6f91" data-selectable-paragraph="">A cat in a large hat made of tropical fruit. --sref https://s.mj.run/O_w9neuqCMI https://s.mj.run/43bAPM-ATlw --sw 999 --v 6.0 </span>

Style reference versus image prompt

I have used the picture of Kitty Miranda as the style reference image with the prompt “a fancy flowered teapot sitting on a table.” Notice how nicely it has picked up the colors, and there is no sign of a cat even though the sref image was just a cat in a fruit hat on a dark wall.

a fancy flowered tea pot sitting on a table. — sref https://s.mj.run/43bAPM-ATlw — v 6.0

a fancy flowered teapot sitting on a table. — sref https://s.mj.run/43bAPM-ATlw — v 6.0

<span id="50dd" data-selectable-paragraph=""><span>a</span> fancy flowered tea pot sitting on <span>a</span> <span>table</span>. <span>--sref</span> https://s.mj.run/<span>43</span>bAPM-ATlw --v <span>6.0</span></span>

Then I used Kitty Miranda as an image prompt with “a fancy flowered teapot sitting on a table” to demonstrate the difference between a style reference versus an image prompt. Notice how we now have a kitty in our teapot.

a fancy flowered tea pot sitting on a table. https://s.mj.run/43bAPM-ATlw — v 6.0

a fancy flowered teapot sitting on a table. https://s.mj.run/43bAPM-ATlw — v 6.0

<span id="ae91" data-selectable-paragraph=""><span>a</span> <span>fancy</span> <span>flowered</span> <span>tea</span> <span>pot</span> <span>sitting</span> <span>on</span> <span>a</span> <span>table</span>. <span>https</span>:</span>

In Conclusion

Wrapping up our exploration of Midjourney’s Style References, I hope you’ve found this deep dive as fascinating as I have. The introduction of the --sref feature marks a significant milestone in the evolution of AI art on Midjourney, offering us an unprecedented level of control over the stylistic consistency of our creations. This tool doesn’t just replicate the aesthetics of a reference image; it integrates its essence into our work, setting a new standard for digital artistry.

Throughout this journey, from grasping the basics to experimenting with complex combinations and weights, every step has been an adventure in discovery. The nuances of how subtle changes can dramatically alter outcomes and the unexpected twists, like the effects of negative weights, have been particularly eye-opening.

This exploration has been more than just a technical exercise; it’s been a creative awakening. Engaging with Style References has expanded my artistic horizons, allowing me to blend inspiration from the world around me into my digital canvases in ways I hadn’t imagined possible just a few months ago. It’s a vivid demonstration of how AI continues to push the boundaries of what’s possible in art, encouraging us to explore, innovate, and create without limits.

I invite you to delve into the vast potential that Style References hold. This feature isn’t just a tool; it’s a gateway to new realms of creativity, waiting to be explored.

Stay tuned for future articles, where I will dive into using this with Niji 6 and bringing in additional parameters to vary our images more.

Thank you for joining me on this exploration. Your curiosity and enthusiasm make these ventures into the new frontiers of AI art all the more rewarding. If this journey has inspired you, I’d love your support with claps (up to 50, no less!). And for more insights into the ever-evolving world of AI and art, make sure to follow me for the latest updates. Let’s continue to push the envelope, explore new possibilities, and redefine what art can be in the age of AI.

a basket full of playful kittens — sref https://s.mj.run/ln_ahmdE43c ::3 https://s.mj.run/Lw_0fjfp0TM ::3 https://s.mj.run/O_w9neuqCMI ::6 — v 6.0 -

a basket full of playful kittens — sref https://s.mj.run/ln_ahmdE43c ::3 https://s.mj.run/Lw_0fjfp0TM ::3 https://s.mj.run/O_w9neuqCMI ::6 — v 6.0 –

prompt