Midjourney Introduces "Style References” To Generate Consistent Style Images

Just days after Midjourney released their highly anticipated V6 model, they launched another exciting feature that lets you generate images with a consistent style. They are calling this brand new feature “Style Reference.”

Style Reference allows you to use the style of an input image and make the AI-generated image results similar to that style.

Let me show you how it works.

How To Use Style Reference

Before diving into the tutorial, make sure that you are a Midjourney subscriber. The paid tier starts at 10 USD per month, and it gives you the benefit of generating images privately on Discord.

Head over to the Midjourney chatbot and bring up the settings menu by typing the keyword “/settings” and selecting the “Midjourney Model V6 [ALPHA]” model.

Next, upload the image you want to copy the style from by clicking on the “+” button and selecting “Upload a File.”

Once uploaded, open the image and copy its address. The image model will use this address as a “style reference” and try to generate something that “matches” its aesthetics.

Now generate a new image using the following format: prompt + "--sref" + URL. Here is an example:

Prompt: a girl wearing a blue sunglasses — sref https://s.mj.run/qaCAwEzNh_0

The AI will generate four images that resemble the style of the reference image. How cool is that? I was really impressed with how close the aesthetics of my results were to the uploaded image.

More Example Results

Prompt: an image of an apple — sref https://s.mj.run/3DbhWRcKpmU

Prompt: an image of a model of a guy wearing green sunglasses — sref https://s.mj.run/1VDX4qCxOuM

Prompt: an image of a scorpion, top view, white background — sref https://s.mj.run/lJjLApGBEv4

You Can Use Multiple Reference Image

Another super cool trick you can do with the Style Reference feature is to use multiple images. Upload two or more images and copy their image addresses. In your prompt, paste the two URLs after the--sref keyword.

The result will be a combination of the aesthetics of the two image inputs.

Prompt: An image of a girl model covered in flowers — sref https://s.mj.run/WffI067qjA4 https://s.mj.run/Oc-DzpfHGcg

A Few Things To Keep In Mind

• Style Reference works for both V6 and Niji V6 (it does not work with V5 and older models)
• Regular image prompts must go before --sref like this/imagine cat ninja ImagePrompt1 ImagePrompt2 --sref stylePrompt1 stylePrompt2
• Set the total strength of the stylization via --sw 100 (100 is default, 0 is off, and 1000 is maximum)
• Style Reference have no direct effect on image prompts, only on jobs that contain at least one text prompt
• Midjourney is likely to update this feature in the next few weeks (and it may change things, so be careful while it’s all in alpha).

More features coming

This isn’t the end of exciting new updates on Midjourney because, according to their office hours on January 31, 2024, there will be more features to be released in the coming weeks.

Midjourney V6

• The current alpha version is getting close to the official V6 beta release
• It is possible for it to be released by end of this week
• The beta model will have another aesthetic improvement

New describe

• Currently under testing with mods
• The new describe feature is a lot better than previous one; it is a lot more accurate, not perfect

Alpha website

• Finally, the image generation on the website will soon open it up to more people who have made 1K images
• The website will also have improvements on the explore page

Niji 6

• Niji will have inpainting and zoom probably next week.
• The new style reference feature will work with Niji 6

Final Thoughts

Overall, I am mind-blown by how fast the Midjourney team is releasing features. It’s almost on a weekly basis. And these are not just minor updates; they are major ones. As a developer, though, I am still waiting for API support. Hopefully, they’ll release it this year.

Considering the tremendous value Midjourney already provides, especially at just $10 per month, any new features feel like the icing on the cake. The team continues to dramatically improve the product, week after week.

I’m excited to see what other surprises are in store. For now, I’m more than satisfied with my subscription and am eagerly experimenting with each update as it drops.

Kudos to the Midjourney team—the pace of progress is incredibly impressive.

‍