// knowledge base
Midjourney Glossary
Terms and parameters, from what you need to get started to the techniques advanced users look for. Every entry in one or two sentences - no fluff. Where it helps, I've added a link to the full article.
Basics
Start here if you're just learning Midjourney.
Prompt
The text description of the image you want the model to generate. It's the only required input - the more precise it is, the more predictable the result.
Minimal prompts win - so why the JSON? →Generation (Job)
A single image-generation request. By default it returns four variants at once, and you pick one to work with further.
Upscale
Enlarging and refining a chosen variant to full resolution. Without this step, you're working with thumbnails.
Variation
Generating variants of a chosen image that keep its overall character but differ in detail.
--ar (Aspect Ratio)
Image proportions, e.g. --ar 16:9 for widescreen or --ar 1:1 for square.
Seed
The starting number that determines a generation's "randomness". The same prompt with the same seed gives a similar result - useful when you want repeatability.
Parameters
Once the basics work and you want more control over the result.
--stylize
How strongly the model interprets your prompt artistically. Lower = closer to a literal description, higher = more creative freedom for the model.
See --stylize in action across 5 non-obvious styles →--chaos
How much variety there is between the four variants in one generation. High --chaos gives more surprising, less predictable sets.
--weird
Experimental unusualness in composition - introduces elements a "normal" Midjourney run wouldn't propose.
--no
Excludes a given element from the image, e.g. --no text, to avoid accidental lettering.
--tile
Generates the image as a repeatable, seamless pattern - useful for textures and wallpapers.
--raw
Dials back Midjourney's default, "smoothed-over" look in favour of a more literal, less stylised interpretation of the description.
Reference Techniques
Control over style and consistency between generations.
Image Prompt
Using a photo as a starting point alongside the text description - paste the image URL at the start of the prompt.
--sref (Style Reference)
Applies a consistent aesthetic from a reference image to a new generation - one of the fastest ways to get a repeatable style across a series.
5 non-obvious styles with ready-to-use prompts →--cref (Character Reference)
Keeps the same character across multiple, different generations - essential for comics, storyboards and any series with a recurring character.
Multi Prompt (::)
Splits a prompt into weighted parts, e.g. horse::2 dragon::1, to control how much each element influences the result.
Permutations {}
Curly braces generate multiple prompt variants at once, e.g. {red, blue} dragon creates two separate generations.
Advanced
Editing, extending the frame and refining finished images.
Inpainting (Vary Region)
Editing a chosen part of an image while keeping the rest of the composition unchanged - you fix one element instead of regenerating everything.
Outpainting (Pan)
Extending the frame beyond the original edges of a generated image, while keeping style and perspective consistent.
Zoom Out
Pulling the camera back from a generated image and filling in the surroundings - a portrait becomes a scene.
Remix Mode
Lets you modify the prompt while creating a variation of an existing image, instead of sticking to the original description exactly.
Personalization (--p)
Tunes the model's style to your individual taste, based on images you've rated previously.
Draft Mode
Faster, cheaper generation of rough drafts before the final, full-quality generation - good for testing composition quickly.