New Style Reference Feature in Midjourney

Everything You Need to Know!

In this newsletter, read about:

  • 🕵️‍♀️ Style Referencing in Midjourney

  • 🗞 News and Top Reads

  • 📌 AI Art Tutorial: Google’s Image Generator

  • 🎨 Featured Artist: Muscat

  • 🖼 AI-Assisted Artwork of the Week

  • 🤓 A Comprehensive Midjourney Guide

🕵️‍♀️ Style Referencing in Midjourney

Midjourney has recently released a new feature – style referencing through the reference image. Now you can add a reference image to your prompt and Midjourney will try to consider only the style of this image, not its content. So, yes, now it’s possible to create an image on a totally different subject, but in the same style as the reference image. It’s not working perfectly yet, but it’s still a very valuable tool.

We’ll discuss some tips and tricks on how to get the best out of this feature, but first – the basics.

How to Use Style References

  • After the text prompt, include the --sref parameter and then, the image URL, e.g. --sref

    https://i.pinimg.com/564x/8d/cf/af/8dcfaf61df73b59d4f2a30f5d7d0d735.jpg

  • To reference multiple images, use multiple URLs, e.g. --sref URL1 URL2 URL3

  • If you want some of the reference images to have a higher impact on the style of your desired image, use relative weights, e.g. --sref URL1 ::3 URL2 ::2 URL3 ::1.

    • The exact numbers don’t matter, but how they relate to each other. For example, the image with a weight of 3 will have a three times higher impact than the image with a weight of 1.

  • To define the total impact of stylization from the style reference images, use the --sw parameter.

    • Varies from 0 to 1000, with 100 being a default value.

Now let’s run a few experiments.

Coloring Page via Style Reference

There are a lot of different tricks to create a coloring page in Midjourney, but actually, there is a variety of possible styles for coloring pages, including cartoon, realistic, mandala, psychedelic, etc.

With the new style referencing feature, you don’t need to search for the right keywords to get the style you want – you can just use the corresponding reference image.

For example, see below how I was able to create two very different coloring pages with Hulk using exactly the same text prompt but changing the style reference image from a simple cartoon style to a psychedelic one.

A coloring page of the Hulk going through the woods. --sref https://i.pinimg.com/564x/46/07/ad/4607ad4e82fe73307babd22c48b18070.jpg --ar 3:4 --chaos 20 --stylize 300 --v 6

And here’s the style reference image on the left and the output image on the right.

A coloring page of the Hulk going through the woods. --sref https://i.pinimg.com/564x/8d/cf/af/8dcfaf61df73b59d4f2a30f5d7d0d735.jpg --ar 3:4 --chaos 20 --stylize 300 --v 6

Looks good! Now it’s time to go a little bit deeper into a more complicated case.

Using Several Style Reference Images

I really liked one of my old images and wanted to create something new keeping the same style.

The first attempts to create a happy loving couple walking through the city failed, because, in addition to the general aesthetics, the output images also had plastic, unrealistic faces (the image above is from an earlier Midjourney version) and gloomy facial expressions. See one example below.

A photo of a loving and very happy couple walking through the empty city. The couple is dressed stylishly in 1960s attire. --sref https://cdn.midjourney.com/c4473d26-103f-4d63-9733-f1143551f6aa/0_3.png --ar 16:9 --sw 30 --v 6

As you can see from the prompt, I also tried to reduce the impact of the reference image by setting the --sw parameter to 30, but it still didn’t help much. So, I decided to add another reference image.

I slightly modified the prompt to mention golden-hour lighting because the second image has distinctly another type of lighting and I wanted to minimize this effect. On my first attempt, both style reference images had the same weight.

A photo of a loving and happy couple walking through the empty city. The couple is dressed stylishly in 1960s attire. It's a golden-hour lighting. --sref https://cdn.midjourney.com/c4473d26-103f-4d63-9733-f1143551f6aa/0_3.png https://cdn.midjourney.com/ef49c367-6195-4d93-86a9-0e8d8dd2bee6/0_2.png --ar 16:9 --sw 30 --v 6

The results were much better, but still too dark.

On my second attempt, I increased the relative weight of the first reference image and got quite good results.

A photo of a loving and happy couple walking through the empty city. The couple is dressed stylishly in 1960s attire. It's a golden-hour lighting. --sref https://cdn.midjourney.com/c4473d26-103f-4d63-9733-f1143551f6aa/0_3.png ::3 https://cdn.midjourney.com/ef49c367-6195-4d93-86a9-0e8d8dd2bee6/0_2.png ::2 --ar 16:9 --sw 30 --v 6

So here are a few parameters to experiment with when style referencing doesn’t work as you want:

  • Adjust the --sw parameter to reduce or increase the impact of style reference images.

  • Add more style reference images. Adjust the weight of each image.

  • Specify in the prompt the style and aesthetic elements that you want to see in the output image.

  • If the output image takes some elements from the reference image that you don’t like, add to the text prompt the details that will replace the undesired elements.

I hope this brief guide will help you use the new style reference feature more effectively. Happy prompting!

🗞 News and Top Reads

  • Google’s Bard chatbot is adding AI image generation to its toolkit.

    • Users can prompt Bard to generate photos using Google’s Imagen 2 text-to-image model.

    • In related news, Google may soon rebrand its AI chatbot Bard as Gemini.

  • Stability AI launches SVD 1.1, an upgraded version of its image-to-video latent diffusion model, Stable Video Diffusion (SVD)

    • This model is optimized to generate short AI videos with better motion and more consistency.

  • Meta is working with industry partners to label AI-generated images on Facebook, Instagram, and Threads.

    • First, they plan to cooperate with other market players on common technical standards for identifying AI content, including video and audio.

    • Next, they will start labeling AI-generated images that users post to their social media when the company can detect the corresponding industry standard indicators. 

📌 AI Art Tutorial: Google’s Image Generator

In this video, Matt explores the image generator finally released by Google. The photorealism is stunning, arguably on par with Midjourney. But the content policy is extremely strict – you cannot even generate an image of an "ugly" cat. Check out the video for more interesting examples.

Muscat is an AI artist creating beautiful imaginary worlds with the help of AI, Photoshop, and a little bit of magic. Check out her magnificent artwork on Instagram @muscatdream.ai.

🖼 AI-Assisted Artwork of the Week

🤓 A Comprehensive Midjourney Guide

To get a link to a comprehensive Midjourney guide, please subscribe to this newsletter. The guide is a dynamic document, which I intend to keep up-to-date with the latest Midjourney updates.

Share Kiki and Mozart

If you enjoy this newsletter and know someone who might also appreciate it, please feel free to share it with them. Let's spread the word about AI art and introduce more people to this fascinating field!

Join the conversation

or to participate.