Midjourney master prompt — anatomy, parameters, and 3 worked examples

The master recipe for any Midjourney image — fill the brackets, then use the guide below to refine:

'[MEDIUM: cinematic photograph / oil painting / editorial illustration / watercolor] of [SUBJECT + ONE KEY DETAIL], [ACTION OR POSE], [ENVIRONMENT + TIME OF DAY], [LIGHTING], [MOOD], [COLOR PALETTE], [COMPOSITION], [STYLE REFERENCE] --ar [16:9 | 3:2 | 4:5 | 1:1] --style raw --stylize [100-500] --chaos [0-30]'

─────────────────────────────────────────
EXAMPLE 1 — Cinematic portrait
'Cinematic photograph of a weathered lighthouse keeper, 70s, salt-and-pepper beard, studying old charts at a wooden desk, raking window light from the left, contemplative and solitary, muted amber and teal palette, rule of thirds, documentary photography --ar 3:2 --style raw --stylize 250'

EXAMPLE 2 — Fantasy concept art
'Digital painting of a wandering desert alchemist, tattered robes hung with brass instruments, crumbled sandstone archive at sunset, dramatic long shadows raking across sand, mysterious and ancient, burnt sienna and turquoise palette, centered wide composition, concept art in the style of Jordan Grimmer --ar 16:9 --style raw --stylize 400'

EXAMPLE 3 — Commercial product shot
'Professional studio photograph of a matte-black ceramic coffee mug, thin wisp of steam curling upward, charcoal seamless backdrop, three-point studio lighting with a warm rim light on the right, clean and premium, dark warm gradient, centered --ar 4:5 --style raw --stylize 150'

─────────────────────────────────────────
THE PROMPT ANATOMY — why each block matters
• MEDIUM first, always. 'Cinematic photograph' vs 'oil painting' vs 'editorial illustration' sets the entire output mode. Without it, Midjourney defaults to a generic digital-art look regardless of everything else you write.
• Subject + one key detail. Lead with the subject and ONE specific detail ('lighthouse keeper studying charts,' not 'old man'). Specificity is what separates memorable images from generic ones.
• Environment + time of day. 'Foggy harbor at dawn' creates completely different light than 'sunny harbor.' Time of day is a lighting instruction in disguise — use it.
• Lighting before mood. Lighting IS mood. 'Soft window light from the left' is more actionable than 'cozy atmosphere.' Always name the light source, direction, and quality.
• Color palette. One 2-word palette ('muted teal and amber,' 'monochrome with deep shadow') controls the entire emotional register. Without it, Midjourney assigns color at random.
• Composition. Without it, Midjourney centers everything. 'Rule of thirds,' 'wide establishing shot,' 'extreme close-up,' 'bird's-eye view' — pick one.

─────────────────────────────────────────
PARAMETERS DECODED
• --style raw: Removes Midjourney's default aesthetic processing — more faithful to your text, less 'AI art' gloss. Use for portraits, product shots, and realism. Remove it for a more curated, stylized look.
• --stylize 0-1000: 0 = follows your prompt literally; 250 = balanced default; 500+ = increasingly artistic and loosely connected to your words. Realism: 100-200. Editorial drama: 250-400. Pure art: 500+.
• --chaos 0-100: 0 = 4 variations of the same idea; 30-50 = meaningfully different interpretations; 80+ = wildly divergent results. Use low chaos to refine a direction, high chaos to explore what's possible.
• --ar (aspect ratio): 16:9 for landscapes/widescreen, 3:2 for photography/editorial, 4:5 for Instagram portrait, 2:3 for book covers/posters, 1:1 for album covers and social squares.

─────────────────────────────────────────
WHAT NOT TO DO
• Don't say 'beautiful,' 'amazing,' or 'perfect' — these words are meaningless to the model. Describe WHAT creates the beauty: the light, the texture, the composition.
• Don't stack more than 2 style references — they compete and cancel each other.
• Don't write 'highly detailed' unless surface texture genuinely matters — it biases toward overworked, cluttered compositions.
• Don't skip the composition block — without it, Midjourney defaults to centered, symmetrical framing for everything.

Tip: use --cref [image URL] to maintain a consistent character across generations; use --sref [image URL] to match a specific style you liked; append --no [terms] to exclude unwanted elements (e.g., '--no text, watermarks, extra limbs'); add 'shot on 85mm f/1.4' for portraits, 'shot on 24mm f/8' for architecture.