Midjourney Prompt Guide: Parameters and Techniques
Good Midjourney output starts with a clear prompt and gets sharpened by a handful of controls that most people never touch. This guide teaches the prompt structure that gives you predictable results, then walks through the controls that actually change the picture: Style Reference, Character Reference, Image Weight, Stylize, Chaos, Creation Modes, Personalization Profiles, and the two editing tools (Vary Region and Remix) that let you fix one part of an image without starting over.
A note on syntax: Midjourney's controls and their exact values change between versions. This guide explains what each control does and names the literal flags where they are documented (--sref, --cref). Always confirm the current syntax and accepted values in Midjourney's official documentation before you build a workflow around them.
Before You Start
Midjourney is subscription-only, and you reach it in one of two places. On Discord you call the bot with the /imagine command and a prompt, and it returns a grid of four images that you can then upscale or vary. In the web editor at midjourney.com you get a single consolidated workspace for creating and editing, with pan, zoom, region editing, and inpainting, and your work syncs with Discord. The older requirement to generate 1,000 images on Discord before reaching the web app has been removed.
How to Structure a Prompt
Before reaching for any parameter, get the words right. Midjourney reads a natural-language description, so the most reliable structure is to move from what the image is to how it should look: subject first, then context and composition, then style or medium. A prompt built this way gives the model a clear anchor and leaves the parameters to do fine-tuning rather than heavy lifting.
Notice there are no parameters in that example yet. That is deliberate. Start with the description alone, generate a grid, and see how close the default output lands. Only then add controls to push the result in a specific direction. The sections below cover those controls in the order you are most likely to need them.
Two habits help here. Keep your first prompt shorter than feels natural, because a tight prompt is easier to diagnose when the output is off. And describe what you want rather than what you do not want, since Midjourney responds far better to positive description than to negation.
Style Reference and Character Reference
The two reference tools solve a problem plain prompts cannot: reusing a specific look or a specific subject across many images. They separate the two things a prompt usually fuses together, which is the look of an image and the thing in it.
Style Reference (--sref)
Style Reference lets you upload an image so Midjourney extracts its palette, texture, and atmosphere and applies that visual style to whatever new subject you prompt. It is how you keep a consistent aesthetic across a set of images without describing that aesthetic in words every time. Point it at a reference and the new generations inherit the mood and color treatment while the subject stays whatever your text prompt says.
Character Reference (--cref)
Character Reference does for subjects what Style Reference does for style: it keeps a consistent character across multiple images. If you are building a sequence where the same person or figure recurs, Character Reference asks Midjourney to carry that character from one generation to the next so it stays recognizable scene to scene, instead of producing a slightly different face each time.
Confirm the syntax: The grounded sources name the --sref and --cref flags and describe what they do, but they do not specify the full syntax, accepted values, or how to combine the two. Check docs.midjourney.com for the current usage before relying on either in a project.
Image Weight: Balancing Prompt and Image
When you supply both a text prompt and an image, Midjourney has to decide which one matters more. Image Weight is the control that sets that balance. Turn it up and the supplied image dominates the result; turn it down and your text prompt takes the lead. It is the dial you reach for when a reference image is either overpowering your prompt or barely registering.
In practice, Image Weight is most useful alongside the reference tools above. If a Style Reference is pulling the composition too far from what you described, lowering the image's influence lets the text reassert itself. If the opposite happens and the model ignores your reference, raising it brings the image back into the picture. The exact parameter and its value range are documented per version, so treat the behavior described here as the concept and verify the current numbers in the docs.
How to think about it: Prompt and image are competing inputs. Image Weight decides which one wins. Adjust it in small steps and regenerate, rather than jumping to an extreme, so you can see which direction the balance is moving.
Stylize and Chaos: Two Very Different Dials
These two parameters get confused constantly because both sound like they make images "more interesting." They do not do the same thing, and knowing the difference saves a lot of wasted generations.
Stylize controls how strongly Midjourney applies its own default aesthetic to your prompt. A low Stylize keeps the result close to a literal reading of your words; a high Stylize lets Midjourney's house style take over, often producing more polished and opinionated images that drift further from a strict interpretation of the prompt. Reach for Stylize when the output feels either too flat or too far from what you asked for.
Chaos (also surfaced as Variety) controls how different the four images in a grid are from one another. Low Chaos gives you four close cousins, useful when you have nearly nailed the look and want minor variations. High Chaos spreads the grid wide, useful early on when you are still exploring directions and want the model to surprise you. It changes the spread across the grid, not the style of any single image.
| Parameter | What It Controls | Turn It Up When |
|---|---|---|
| Stylize | Strength of Midjourney's default aesthetic | Output feels too literal or flat |
| Chaos (Variety) | How different the 4 grid images are | You are exploring and want range |
| Image Weight | Prompt vs supplied image influence | A reference image is too weak |
Exact value ranges and defaults for these parameters are version-specific. Confirm them in Midjourney's documentation.
Creation Modes and Personalization Profiles
Two settings shape your output before you even write a prompt: which Creation Mode you are in, and whether your Personalization Profile is active.
Creation Modes are selectable modes that change how Midjourney generates images. They are a setting you choose for a session rather than a flag you add to a single prompt, so it is worth knowing which mode you are in before you wonder why two similar prompts produced different kinds of results.
Personalization Profiles tune Midjourney toward your own preferences. A profile is built from your choices and then biases future generations toward the kinds of images you have favored. As noted earlier, the current V8.1 release requires the Global V7/V8 Personalization Profile to be enabled, so personalization is not a fringe feature here; it is part of the standard setup. Midjourney publishes an official "Get Better Images with Personalization" tutorial that walks through building one.
Version context: V8.1, released in alpha on April 14, 2026, is the current stable version. It keeps the V7 aesthetic and adds HD images, Image Prompts, a Prompt Shortener, and an updated Describe tool. If a tutorial or forum post references V7 as "the latest," it is out of date. Confirm the current version on Midjourney's help center.
Editing Output: Vary (Region) and Remix
Once you have a result that is almost right, you rarely need to start over. Two tools let you change part of an image while keeping the rest.
Vary (Region) is region editing, also called inpainting. You select an area of an existing image and have Midjourney regenerate only that selection, leaving everything outside it untouched. It is how you fix a single awkward hand, swap an object, or clean up one corner without rerolling the whole composition. Region editing has been part of Midjourney since V5.2.
Remix changes the prompt or settings at the moment you vary an image, so your edits combine new instructions with the existing composition. Where Vary (Region) targets where the change happens, Remix targets what the change is: you keep the structure of the image you liked and feed in a revised prompt to steer the variation. The two are often used together, selecting a region and then remixing the prompt for just that area.
Honest Limits of This Guide
This guide is built from a focused set of sources that describe what Midjourney's features do, not a complete parameter manual. A few honest caveats are worth stating plainly so you do not get tripped up.
Troubleshooting Common Prompt Problems
These are the issues most people hit once they move past their first few prompts, with the control that usually fixes each one.