The Complete Guide to Midjourney Parameters: --ar, --s, --v, --c, --no

In Midjourney, a prompt consists of the image description and the parameters. Parameters start with two dashes, go at the end of the prompt, and are separated from the text by a space, for example: a cat on the roof --ar 16:9. Parameters control how the image is generated rather than what it contains, including the aspect ratio, style tendency, model version, how different the four initial images are, and which elements to exclude. The same description with different parameters can produce very different results. Change only one parameter at a time, compare the four images, and your intuition will build up gradually.

--ar stands for aspect ratio and controls the proportion between the width and height of the generated image; it is the most frequently used parameter. Commonly used values include: 1:1 square for avatars, icons and stickers; 3:2 and 4:3, close to traditional photo proportions, for landscapes and everyday scenes; 16:9 widescreen for wallpapers, banners and covers; 4:5 and 3:4 vertical for social media portraits; and 9:16 full-screen vertical for phones. Without --ar, Midjourney usually defaults to a square or near-square image.

How to tune it: choose a wide ratio when the scene spreads horizontally, and a vertical ratio when highlighting a single person or object. Complete examples: for a landscape banner, try misty mountains at sunrise --ar 16:9; for a vertical portrait, try close-up portrait of a woman, soft window light --ar 4:5; for a square avatar, try cute robot mascot, flat illustration --ar 1:1. The more extreme the ratio, the more you should describe the composition of the main subject in advance, so important content is not cropped out.

--s stands for stylize, with a common range of 0 to 1000, controlling how strongly Midjourney's built-in default aesthetics influence the image. At low values the picture stays closer to the literal description of the prompt, leaving the model little room of its own; at high values the model adds more of its preferred composition, colors and artistic treatment, which usually makes the picture more polished but may drift away from your description. It is a trade-off between control and visual appeal.

General advice: use low values for subjects that need precise content control, such as product shots, for example white ceramic mug on wooden table, studio lighting --s 50; use higher values for illustrations and concept scenes where atmosphere matters more, for example floating city above clouds, dreamy atmosphere --s 750. When unsure, start near the default value and compare with small adjustments.

--v stands for version and specifies which generation of the model to use, written for example as --v 7, subject to the officially current version. Newer versions usually bring better image quality, lighting and understanding of complex prompts, but the aesthetic also shifts between versions, and the same prompt can look noticeably different across them. Without --v, the officially recommended current version is used by default.

How to tune it: in most cases use the latest version. When a style worked better in an older version, or you want to reproduce a past result, specify the older version explicitly. Examples: for photorealistic subjects try old bookstore interior, dust in sunbeams --v 7; for retro posters try retro sci-fi poster, 1980s style --v 6. After switching versions, re-tune the prompt instead of reusing the old parameters.

--c stands for chaos, with a range of 0 to 100, controlling how different the four initial images in one job are. At low values they share similar compositions, making it easy to compare and pick details along one direction; at high values they differ clearly in composition and style, which suits early-stage divergence and unexpected directions.

General advice: use low values when you already know what you want, for example minimalist logo of a fox, geometric --c 0; use higher values for brainstorming and inspiration, for example surreal dream landscape, unexpected elements --c 80. For everyday image making, staying in the low-to-medium range avoids excessive quality swings.

--no excludes elements you do not want. It goes at the end of the prompt, with multiple elements separated by commas, for example --no text, watermark. It is often used to remove text, watermarks, people, unwanted objects or specific colors for a cleaner result. Note that it is not a hard deletion; it only lowers the probability of these elements appearing, and they may still show up in complex scenes.

Examples: for a clean natural landscape, try calm lake reflection at dusk --no people, buildings; for a food still life, try fresh strawberry cake on plate --no hands, text. If the result is still not ideal, the more effective fix is usually a more specific positive description rather than more parameters.

Parameters can be freely combined, for example sunset over desert dunes --ar 16:9 --s 250 --c 30 --no text. A recommended order: set the canvas with --ar, decide stylization strength with --s, add --c while exploring, remove unwanted elements with --no, then lock in a satisfying set as the default preset for similar subjects.

For beginners, the most effective practice is controlling variables: change one parameter at a time and compare the four images, and intuition builds up quickly. These five parameters already cover most everyday situations; once comfortable, it is not too late to explore other advanced parameters in the official documentation.