Midjourney Complete Prompting Handbook

Midjourney has an opinionated, aesthetic-first style — and a prompt language all its own. Learn the structure pro users follow, the parameters that actually matter, and the patterns that turn one-off images into a coherent visual identity.

Note: Midjourney updates models and parameters frequently. Always confirm flag syntax in the official documentation.

1. Introduction

Midjourney is famous for outputs that look beautiful by default. The trade-off is that the model is heavily styled — you are not so much commissioning a literal image as collaborating with a model that already has strong artistic preferences. To get specific results, you have to learn its prompt grammar and lean into its strengths instead of fighting them.

This handbook walks through the prompt anatomy, parameter syntax, the powerful reference flags --sref and --cref, and the workflow patterns that keep a project visually consistent across many images.

2. The Midjourney Prompt Anatomy

A strong Midjourney prompt is rarely a sentence. It is a comma-separated stack of visual signals, each one contributing to the final image. The order matters: tokens earlier in the prompt carry more weight.

Beginners typically include only the subject. Pros include all five layers in nearly every prompt — that single habit is what makes their output look "designed" instead of "generated".

3. Subject, Style, Composition, Lighting

Each layer has its own vocabulary worth learning.

• Subject: who or what, what they are doing, where. Be specific. "a woman reading" is weaker than "a librarian in her late 60s reading a thick hardback on a bench".
• Style: medium (oil painting, ink wash, 35mm film, isometric 3D render), and one or two artistic references (a movement, an era, a director's cinematography). Avoid mimicking living artists.
• Composition: shot type (extreme close-up, full body, wide), camera angle (low angle, top-down, Dutch tilt), and framing (rule of thirds, symmetrical, negative space).
• Lighting: direction (back-lit, side-lit), quality (soft, harsh), colour temperature (warm amber, cool blue) and time-of-day cues. Lighting often does more for mood than the subject itself.

a lone fisherman casting a net at dawn, weathered wooden boat,
oil painting, Edward Hopper influence,
wide shot, low horizon line, figure positioned right of centre,
soft amber morning light, thin sea fog, long shadows,
--ar 3:2 --style raw --stylize 250

4. Parameters That Actually Matter

Midjourney parameters are flags appended to your prompt. A handful matter daily; the rest are situational.

• --ar 16:9, --ar 3:2, --ar 9:16 — aspect ratio. The single most important parameter.
• --stylize 0–1000 — how aggressively Midjourney imposes its house aesthetic. Lower (50–100) = more literal. Higher (500–1000) = more stylised, often more beautiful but less accurate.
• --style raw — turns down the default beautification. Useful for photographic realism and design mockups.
• --chaos 0–100 — variation between the four returned options. Higher = more wildcard. Useful in early exploration.
• --no [thing] — exclude something the model keeps adding. --no text, watermark is a common cleanup.
• --seed [number] — reproduce the same starting noise. Lets you iterate on a variant while keeping the base composition stable.

5. Style and Character Reference: --sref and --cref

The reference flags are Midjourney's killer feature for projects. --sref applies the visual style of a reference image to a new prompt. --cref tries to preserve a character across images. Both accept image URLs.

vintage botanical illustration of a mango tree in fruit,
detailed leaves, hand-drawn pencil lines, paper texture,
--ar 4:5 --sref https://your-image-url.png --sw 150

For a brand or a children's book project, generate or pick one "anchor" image, then use --sref on every subsequent generation. The result is a coherent set instead of a random assortment.

6. Workflow Patterns Pro Users Follow

• Explore, then commit. Use --chaos 30 on early prompts to see range. Once a direction feels right, lower chaos and start tightening.
• Upscale and vary. Don't just upscale your favourite — use the "Vary (Subtle)" and "Vary (Strong)" buttons to push promising images further before committing.
• Build a style library. Save 3–5 anchor images per project. Reuse them with --sref for visual consistency.
• Prompt by reference, not adjectives. "1970s Polaroid" is more powerful than five style adjectives. Concrete cultural references work better than abstract aesthetic words.
• Iterate in pairs. Change only one variable at a time (style, lighting, aspect ratio). Otherwise you cannot tell what caused an improvement.

7. Practice Exercises

8. Key Takeaways