Midjourney Style Creator: Train AI for Your Visual Style

👁 115 reads

Style Creator is a Midjourney tool that lets you build your own repeatable visual style without writing long descriptions. It shows you grids of images, you click the ones that match your aesthetic, and the tool creates a unique --sref code you can use in every subsequent project. For branding, illustration, and anything requiring visual consistency, it’s one of the most convenient features in Midjourney.

We updated this article on September 30, 2026 to reflect Midjourney V8.2, which has been the default version since July 24, 2026. The most important thing we found in the documentation: Style Creator itself still generates preview images in version 7, but the style codes it produces also work in V8.1 and V8.2. Below we describe what this means in practice. For the full picture of how much Midjourney costs and what V8.2 is, see our Midjourney review.

The tool is only available on the midjourney.com website - it’s not on Discord. After several rounds of selection, you get a style code that you can attach to any prompt and generate images in the same aesthetic, regardless of content.

What Style Creator Is and How It Works in the V8 Era

Style Creator is a tool available in the left navigation menu on the Midjourney website. Its purpose: let you define a visual style by choosing images, not words. The system creates custom style reference codes (--sref) that you can use repeatedly in prompts. Each code is a saved set of aesthetic characteristics - color palette, textures, mood, lighting treatment.

An important detail from the documentation: images generated in Style Creator are created in Midjourney version 7. If you enter the tool with a prompt that has parameters incompatible with V7, you may see an error - you’ll need to remove them. The style codes themselves are compatible with V8.1 and V8.2: in the compatibility table on the Version page, Style Reference codes, Style Reference images, and the --sw parameter all show “works” for V8.

Regarding the Style Reference system version: Midjourney changed it in V7 on June 16, 2025. In version 7, --sv 6 is the default, and --sv 4 is the old sref model from V7 before that date. Codes from before the change look different today, so for old codes add --sv 4 or switch to V6. The documentation describes --sv versions only for V6 and V7 - it doesn’t provide such a choice for V8.1 and V8.2.

Key feature: Style Creator learns from both what you select and what you don’t select. With each round, the styles shown get closer to your vision. Each round produces a new set of preview images with its own --sref code. At the end of the session, you have several codes to choose from - each corresponds to a different stage of refinement.

How to Launch Style Creator - Requirements and First Prompt

You need an active Midjourney plan (starting from Basic at $10 USD per month - there’s no free plan) and access to midjourney.com. The Style Creator documentation doesn’t list other requirements. In an earlier version of this text, we wrote about needing to rate about 200 image pairs - that applied to unlocking personalization in V7, and Midjourney has already replaced pair rating with selecting images from a grid. Personalization is useful if you want to combine it with style codes, but it’s not described as a requirement for Style Creator.

There are two paths to the tool. First: click Style Creator in the left menu on midjourney.com and enter a base prompt. Second: on the Create page, hover over the text of a prompt you already have and click the Style Creator button. This is important: the prompt in Style Creator only serves to generate preview images after each round, but doesn’t control the tool itself. You can enter anything - “portrait of a woman,” “science fiction landscape,” “minimalist poster.” Images in the grids may not match the prompt thematically and that’s normal - you’re choosing style, not content.

The documentation advises: a simple prompt best shows what the style does. A complex prompt with additional parameters mixes its own details with the new code, so the same code used later with a different prompt may look different.

If you want to speed up previews and save GPU time, add --draft to the prompt. Because previews are created in V7, Draft Mode from version 7 works here: according to Midjourney, it costs half the standard GPU time and renders about 10 times faster. Quality is lower, but it’s sufficient for evaluating style. Remember that all preview images in Style Creator consume your GPU time - in the Basic plan you have 200 minutes per month, and an additional hour costs $4 USD.

The Style Creation Process - From Chaos to Stable Code in 5-15 Rounds

After entering a prompt, Style Creator shows a grid of images in various styles. You scroll through the page and click those whose aesthetic appeals to you. Focus exclusively on style - colors, texture, mood - not on what’s in the image. As you scroll and select, the tool periodically regenerates your prompt into new preview images with their own style code. Each such set is one round. Hovering over a thumbnail shows you three images in a larger preview, and clicking pins them in the side panel.

Early rounds can be chaotic. Styles seem random and jump between aesthetics - according to the documentation, this is normal because the tool is still narrowing down your taste. Most styles stabilize after 5-10 rounds, rounds 10-15 add more detail, and after round 15 changes are small and subtle.

If no style in the grid appeals to you, just keep scrolling - skipping images doesn’t affect style development. It’s worth enabling keyboard shortcuts (the shortcuts icon in the interface) because they speed up selection. After hovering over a sample style, you can also immediately try it out, see details, or save it to your favorite styles on the Explore page.

When you’re satisfied with the previews, click X or “End Session.” Warning: you can’t reopen a session after closing it, so make sure you’re finished. All preview images and style codes from the session are saved on the Create page.

How to Use Generated --sref Codes in V8.2

After ending the session, on the Create page you’ll see sets of preview images - each set is one round and has its own style code. Click the code and it will be added to the Imagine bar. Now you can enter any prompt and use that style.

Example: your code is --sref 4802911573. You enter “cat sitting on a roof --sref 4802911573” and you get a cat in the style chosen in the session. You change the prompt to “futuristic city at night --sref 4802911573” - the city gets the same color scheme, mood, and textures.

Here’s where version differences come in. Previews in Style Creator were created in V7, and if your default version is V8.2, the code will be used by the newer model. The documentation doesn’t promise that the same code will give an identical effect in V8.2 as in the preview. Before a longer series, we recommend testing the code on several prompts with --v 8.2. If you want exactly the look from the preview, add --v 7 - version 7 is still available.

You can combine style codes. If you use --sref 4802911573 7031655206, Midjourney will blend both styles. You set the strength of the style’s influence with the --sw parameter (style weight): range 0-1000, default 100. A higher value (e.g., --sw 500) strengthens the style, a lower one (--sw 50) weakens it in favor of the text prompt. According to the documentation, in V7 the --sw parameter has more impact with style codes than with reference images.

Old --sref codes discovered before June 16, 2025 require adding --sv 4 (which means working in V7) or switching to V6. Example: “portrait --sref 287298887 --v 7 --sv 4” will use the old algorithm, and “portrait --sref 287298887 --v 7” the new one.

Advanced Tricks - Personalization, Moodboards, and Combining Styles

You can combine style codes with personalization. The --p parameter adds your aesthetic profile to the prompt, and --sref adds a specific project style. In V8.1 and V8.2, your global personalization profile from V7 works, and for V8 you can create additional profiles (they don’t work backward in V7). You set personalization strength with the --stylize parameter (0-1000, default 100).

You can also start from an existing --sref code. The documentation allows you to enter a style description or an existing style code at the start in Style Creator. However, you need to know how this works: the new code from the session doesn’t replace the old one, it’s added alongside it. Codes don’t merge - both remain active, and previews are created with both codes at once. To recreate the look from Style Creator, use both codes together in subsequent prompts. The same happens when you open Style Creator again from an image from a previous session: each new session adds another code.

Moodboards are a second path to your own style: you upload a set of images and Midjourney extracts their common aesthetic. Moodboards work in V8.1 and V8.2, and according to Midjourney, in V8.1 moodboards and sref became “super stable.” The --sw parameter doesn’t work with moodboards, but codes from Style Creator work with them without problems.

On the Explore page, in the Styles tab, there’s also Style Explorer: you browse ready-made style codes (random or popular in the community), search for them by keyword, e.g., “anime” or “photographic,” and save favorites. In the test version of the service (alpha.midjourney.com), Midjourney added “Live previews” in September 2026, which shows the current prompt in different styles, and a “Change Style” function in the image preview. These are experiments, so they may still change.

The Midjourney community talks about over 4 billion existing style codes. Midjourney doesn’t officially provide this number, so treat it as an estimate. Codes from Style Creator are created during your session - you don’t choose them from a ready-made list.

What It’s Good For - Use Cases

Branding and visual consistency. If you’re running a project requiring one aesthetic across many images - social media, presentations, portfolio - Style Creator is a good tool. You create a code once, then use it in all prompts. Consistency is high, though not absolute: with very different subjects (people, objects, landscapes) the style still shifts somewhat.

Commercial illustration. Designers who need a series of images in one style (e.g., for a book, game, application) can create an --sref code tailored to the client’s brief. Instead of long descriptions like “pastel colors, soft light, hand-drawn texture,” a code is enough.

Exploring new aesthetics. Don’t know what style suits you? Style Creator works well as an experiment. You click images intuitively, and the tool shows you styles you hadn’t thought of - including ones that are hard to describe in words.

Character and world consistency. Style Creator doesn’t transfer characters - a different tool does that (more on that below) - but it helps maintain the same visual atmosphere throughout a story. In worldbuilding, comics, or concept art, an --sref code ensures all scenes have a similar mood.

Problems and Limitations - What Style Creator Can’t Do

The tool has limitations. Early rounds can show styles completely divergent from expectations, and first sessions can be frustrating until you understand the logic of choosing style instead of content.

Style Creator doesn’t copy specific objects or characters - only style. If you need the same character in different scenes, in V8.1 and V8.2 the Edit Model serves this purpose: it accepts up to four reference images and replaced Omni Reference and Character Reference (--cref). Omni Reference (--oref) now only works in V7. Edit Model can be combined with style codes, though Midjourney advises also describing the style in words in the prompt.

Previews are created in V7, not in the default V8.2. This means what you see in the session is a guide, not a guarantee of appearance in the newer model. Before a series, test the code in --v 8.2.

You can’t save a session and return to it later. If you close the window before finishing, you can’t return to that session (codes from completed rounds remain on the Create page). This is a problem if the session runs long, for example 15 rounds or more.

After round 15, changes are small and subtle according to the documentation. If you’re not a perfectionist, 10 rounds usually suffice, and each round is additional GPU minutes.

Comparison with Other Style Control Methods in Midjourney V8.2

Midjourney V8.2 has several ways to control style: Style Creator (creates custom --sref codes), Style Reference with images (you upload an image as a style template), style codes from Style Explorer, moodboards (a set of images as a template), and personalization (your aesthetic profile). For transferring characters and objects and for “repainting” an image in a different style, Edit Model serves this purpose, which has the former Retexture tool built in. We described this more broadly in the guide Style Reference in Midjourney.

Style Creator versus reference image: --sref with an image URL uses a specific image as a template. The result depends on how Midjourney interprets it and can be less predictable. Style Creator gives a numeric code that works repeatably in the same model version - that’s why for a series it’s worth sticking to one version.

Style Creator versus personalization: personalization (--p) is an aesthetic profile built from your image choices and applies to all generations where you enable it. Style Creator is a specific style for a specific project. You can have a profile that likes realistic portraits and simultaneously an --sref code for abstract illustration. Together, --p and --sref give your style adapted to your taste.

Style Creator versus long prompts: instead of writing “watercolor painting, soft pastel colors, impressionist brushstrokes, dreamy atmosphere, vintage paper texture,” you use one code. Shorter, repeatable, and with less risk that Midjourney will misunderstand a word.

Is It Worth Investing Time in Style Creator?

If you work on projects requiring visual consistency - in our opinion, yes. A session of a dozen or so rounds takes some time and GPU minutes, but the finished code then saves iterating on text descriptions for each image.

For experimenters and artists seeking new aesthetics - also yes. Style Creator is a tool for discovery, not just control.

For people creating individual images without the need for consistency - rather unnecessary. If each image is a separate project, a description or reference image is enough.

Midjourney strongly emphasizes personalization and style control. With competitors, style is usually set by a reference image in the prompt or, in open models like FLUX, by training your own LoRA adapter, which requires more work. Just remember two things: Style Creator works on previews from V7, and images in Midjourney are public by default - you’ll only hide them with Stealth mode in the Pro plan at $60 USD per month. We described plan details and evaluation of the entire tool in the Midjourney review.

Kinetyra