Using the Midjourney /tune Command

Midjourney Style Tuner

In previous article we explained possibilities of new Midjourney /tune command in general. Let’s go deep in using the Midjourney Style Tuner.

How to use Midjourney Style Tuner

Launch our new Style Tuner tool with the /tune <prompt> command!

To load the Style Tuner:

  1. Use /tune followed by a prompt, like this: /tune splashes of mercury
  2. You will have an opportunity to configure the Style Tuner to have 16, 32, 64, or 128 visual directions to choose from.
  3. Agree to the GPU cost and be prepared to wait a few minutes for your Style Tuner to be populated with sample images based on the prompt you supplied.

Using the Style Tuner, you can explore billions of --styles by selecting from among different visual examples, each of which represents a behind-the-scenes configuration of Midjourney that will pull the style in one direction or another. (Your prompt is NOT part of the --style, it’s just responsible for the sample images displayed by the Style Tuner.)

New to --style? Learn more here.

Did you know that --style is deeper than a visual aesthetic or vibe? Midjourney --style will effect composition, coherency, medium, mood, subject, and much more.

Choose between visual directions to generate a --style code, copy it from the Style Tuner, then come back to Discord and use “--style <code>” at the end of a prompt to apply it.

What is the Style Tuner?

The Style Tuner is a VISUAL way to explore --styles! When you use /tune you will get a link to a web page with sample images based on your prompt. (The page opens in a browser window, not in Discord.)

When you use /tune you can configure the Style Tuner to show you 16, 32, 64, or 128 rows or pairs of sample images representing different visual directions. The prompt does not inform the resulting --style in any way.

You can choose between two views for browsing styles:

  • “Pair” view. Each row shows two visual directions. Pick whether you want the left or the right style to blend into the final aesthetic you’re creating, or leave the row in its neutral middle selection.
  • “Grid” view. Each image you click pulls the aesthetic you’re creating in that direction, and leaves another aesthetic direction behind. Selecting neither direction is the neutral position.

What is a --style code?

Style codes look like this: 39BKrrsVsbfxJl6

When you apply a --style you are reconfiguring Midjourney for vibe, aesthetics, coherency, compositional tendencies, and more. (David says: “It’s like changing Midjourney’s personality.”).

Use “--style <code>” at the end of a prompt to apply the style. (Only compatible with current version of Midjourney). As always, --style replaces Midjourney’s default house style. You can also use “--style random” to apply a random style.

You can blend multiple styles together by separating codes with dashes, like this: code-code-code.

The --style parameter requires a text prompt (you can’t give it just URLs) and is not compatible with --weird or --niji.

When applying a style to a prompt, you can adjust the strength of the --style with --s between 20 - 1000. Values below 100 will decrease the contribution of the --style, in favor of the prompt. Values over 100 will increase the contribution of the --style and may also help with stability/coherence.

Things to keep in mind

There are 3^128 total possible --styles! The Style Tuner helps you explore this vast library.

If your --style code needs troubleshooting, start exploring higher values of --s (250, 500, 1000) to increase your style’s influence and possibly even boost coherency.

If the prompt to which you apply the --style contains its own aesthetic that style will blend/compete with “--style“.

Creating a Style Tuner costs GPU minutes. You’ll get an estimate when you use /tune. After that, anyone with the URL can play at no cost to you or to them.

The images made by /style will NOT be visible in your web gallery. If you know how to use the browser “Inspect” feature, you can get the job_id and use /show job_id to bring those images into your workspace.

If you are on a mobile device, we recommend you open the Style Tuner on your mobile web browser rather than trying to use the Style Tuner on Discord’s instance of the browser.

What can I do with --style?

You can pin a style to keep using it without re-typing it. Type /settings, use the [📌 Sticky Style] button, and the last style you used will stay in play until you replace it.

You can save and organize styles for later use. You can save your codes using --style Parameter. For example “/prefer option set kawaii-line-art --style 5sm680fRODCj

Also, you can share styles. Any member of Midjourney can use any style code.

Blending multiple --styles.

You can blend multiple styles together by separating codes with dashes, like this: --style codeA-codeB-codeC. You can blend in “--style raw” and “--style random” by including the word raw or random as a code.

If your codes are all exactly the same, it’s as if you’ve used the code just once: codeA-codeA-codeA is the same as codeA. If your styles are different but some of them repeat, you’re weighting the repeated style a little bit: codeA-codeA-codeA-codeB means codeA will contribute more than codeB.

There’s no way to “negatively” weight a style.

Using --style random.

Using --style random replaces ‘random‘ with a style code. You can roughly control the variety of styles explored by --style random by specifying the number of possible visual directions in play, and what percentage of those visual directions will be active.

The number of visual directions possible is represented by the number of rows in the Style Tuner: 16, 32, 64, or 128. Let’s call this length.

The number of visual directions we’ve activated is usually determined by our responses to the rows or pairs in the Style Tuner. But, in the case of random, we must manually specify what percentage of the possible visual directions we want to activate: e.g. all of them (100), half of them (50), a quarter of them (25), etc. We can call this percentage.

So we can use random-length-percentage to control the variety of styles to be explored. For example, random-128-100 means 128 visual directions are possible and all are active, and random-16-25 means only 16 directions are possible, and 25% are active.

The more visual directions in play, the larger the pool of possible styles to explore. The higher the percentage of activated visual directions, the less visual difference we can expect between those styles (because so many directions are defined, closing in on a look). So random-128-100 will likely produce styles that are visual cousins to each other, but random-128-5 is more likely to produce styles with more obvious visual differences.

Use --repeat and/or {parameters} with --style random-length-percentage for quick style exploration, especially in /relax mode, where it’s free.

Using the Style Tuner

The /tune command and sample images

You can’t influence what the visual directions are doing. There are 128 of them and they appear in a fixed order. The only thing you can customize are the sample images, which come from your prompt.

When you use /tune <prompt> we recommend using prompts with no parameters.

That said, --ar, --chaos, and --tile will be carried into the copy-paste output for you. (Check out the bottom of the Style Tuner.)

Weights and image references in /tune <prompt> will also only effect the sample images, and not the --style.

When running /tune you will be able to specify that the sample images should be rendered in raw mode, which is helpful if you plan to use the resulting --style in raw mode.

Reminder: Nothing about the prompt will be baked into the resulting --style. The prompt and parameters are used only to create sample images to help you assess what the visual directions are doing. The effect is that some --styles may appear to perform best with very similar prompts.

The Style Tuner is the same tool no matter what sample images it displays. The samples are tailored to the prompt you provide to help you picture how that particular prompt would be influenced by the visual direction you’re selecting. When you share a Style Studio URL, you are sharing an instance of the Style Studio tool with your prompt’s sample images.

The potential variety of styles

You can roughly control the variety of styles produced by the Style Tuner by specifying the number of possible visual directions in play (16, 32, 64, or 128) and then choosing among them.

The number of visual directions possible is represented by the number of rows or pairs in the Style Tuner: 16, 32, 64, or 128. The more visual directions to choose from, the larger the pool of possible styles.

Which of these visual directions we’re blending together is determined by the responses to the rows or pairs in the Style Tuner.

Selections are blended together.

If you blend a high number of visual directions together (by choosing lots of them), you might not be able to perceive exactly how your choices are playing out in the resulting style.

Selecting only a few visual directions is more likely to produce a style you can visually correlate to your choices.

Free to use Preconfigured Style Tuners

Don’t want to spend any GPU minutes to try the Style Tuner? Select one of these pre-configured Style Tuners to create styles for photographs, fantasy art, architecture, or landscapes.

  1. Styling Photographs: https://tuner.midjourney.com/dEUewPi,
  2. Styling Fantasy Art: https://tuner.midjourney.com/FblOgmu,
  3. Styling Architecture: https://tuner.midjourney.com/8HeI1uR,
  4. Styling Landscapes: https://tuner.midjourney.com/99xWBJP.

Here’s what a (cropped) Style Tuner looks like, by the way:

FAQ

What’s the difference between the pairs view and the grid view?

It’s just a cosmetic difference.

  • “Pair” view. Each row shows two visual directions. Pick whether you want the left or the right style to blend into the final aesthetic you’re creating, or leave the row in its neutral middle selection.
  • “Grid” view. Each image you click pulls the aesthetic you’re creating in that direction, and leaves another aesthetic direction behind. Selecting neither direction is the neutral position.

How do my selections contribute to the final --style?

This is not something we know in technical detail. You can think of the Style Tuner as a cauldron. You’re adding ingredients. Magical alchemy happens to create a --style.

Read other related articles:


Posted

getimg-ad