Midjourney: Advanced Style Exploration

How to recycle and reuse the unwanted Style Reference codes to create stunning images

Feb 14, 2025

∙ Paid

You can tell the bot not to create ugly images.

If you use --sref random to mine Sref codes, you'll quickly notice that most of them aren't visually appealing. The failure rate is high. It often takes many runs to find a usable code, and even more iterations to discover a unique, beautiful Sref code that make you feel like winning a small lottery.

The question is, what should we do with the unwanted Sref codes that we hardly look at them twice? They will be forgotten and abandoned, just like many other images that did not make the cut during the selection process.

But there is another way.

We could “recycle” and put them to good use.

Style Exploration Hack: A Discord-only Technique

Clarinet from Midjourney shared this new technique called “Style Exploration” that uses multi-prompts on Discord. It's a hack that tricks the MJ bot into executing a prompt that would otherwise impossible.

Multi-prompting with weight assignment is an advanced Midjourney technique. You can read more about Multi-prompts here.

At its most basic, a multi-prompt can be used to remove an element from an image (for example, a dog). When the multi-prompt is assigned a negative weight, the effect is similar to that of the --no parameter. For instance, dog::-0.5 is a multi-prompt that generates images without a dog.

(There are a lot more things to talk about in Multi-prompt. I will cover that topic another day).

The Style Exploration technique applies the same principle of negative prompting. Instead of removing an element, it instructs the Midjourney bot to move away from the aesthetic styles you have chosen. For example, if you prefer bright and colorful images, you can instruct the bot to avoid creating dark images with black and gloomy colors.

In other words, you are telling the bot to avoid the aesthetic spaces that generate those dark and gloomy images.

It's fine if the above doesn't make sense to you. That’s the theoretical part. You can ignore all of that and use the provided template and examples below to create nice images. Worry less, create more, and have fun :-P

Here's the prompt template I adapted from the original technique to make it easier to remember with improved results.

Prompt template (Only works in Discord)

/imagine prompt: [Prompt] --sref URL1::-.5 URL2::-.5 URL3::-.501 --sw 100 --ar 16:9 --v 6.1

Replace [Prompt] with your prompt e.g. a lady and a cat.

URL1, URL2, and URL3 are links to the reference images that you DO NOT LIKE. They have the aesthetic styles that you want to avoid. You can include more than three links.

One of the URLs weight must have at least 3-decimal point like this (-.501) to make it work.

The numbers after the double semi-colons are the weights of the multiple prompts. You can keep them as they are.

The Style Weight is set to a default value as a placeholder for future adjustments. (Read the benefits of using a default placeholder here)

The main ingredients for this technique are the URLs of the images you dislike. Find those ugly and unwanted Sref codes that you've abandoned. Copy the links to those images. Then, you use the multi-prompt technique to explicitly tell the bot to stay away from those aesthetics.

Example: Make it colorful

Original image and the links to the ugly images to avoid.

The original image generated using the prompt: /imagine prompt: a lady and a cat --ar 16:9 --v 6.1 —> nothing fancy here. Just an ordinary image. But we can improve it using the Style Exploration technique.
Link to a dark image that I don’t like: https://cdn.midjourney.com/fe642c61-8901-4914-9de9-3cdb4d902f9c/0_0.png
Link to an ugly image that I don’t like: https://cdn.midjourney.com/b525595b-f039-408b-9f29-1c51a920906d/0_1.png
Link to an image where I don’t like the color scheme: https://cdn.midjourney.com/9b1205d0-a9c5-43ed-8852-e57914e84a1b/0_0.png

With all of the ingredients set up, we can now use them in the prompt template to get rid of the dark, ugly, and undesirable color attributes.

The result

/imagine prompt: a lady and a cat --sref https://s.mj.run/i9XmgafleXc ::-0.5 https://s.mj.run/5jTSJVAaUFM ::-0.5 https://s.mj.run/dPxxmip-iEo ::-0.501 --ar 16:9 --sw 100 --v 6.1

So, no more dark and gloomy vibes. By instructing the bot to avoid those aesthetic styles, the generated images become bright and colorful.

Interesting, huh?

We have found a way to make use of those unwanted Sref codes for image creation.

Finally, ugly Sref codes can have their day too!

Note: You can use the same technique with Personalization codes. To keep things simple, I will not go into detail because the Personalization feature may be upgraded very soon in V7, and the technique may be obsolete by the time you read it.