The Complete Guide to Writing Midjourney V7 Prompts (With Examples)

Use five ordered slots, comma-separated:

1. Subject — who or what is in the image.
2. Environment — where and when.
3. Style — art movement, medium, or reference.
4. Lighting & mood — how it feels.
5. Technical — camera, lens, quality words, parameters.

Example: "elderly fisherman mending a net, weathered wooden dock at dawn, hyperrealistic oil painting, warm rim light and cool fog, 85mm portrait lens, shallow depth of field --ar 3:2 --stylize 400"

• --ar 16:9 Aspect ratio. Use 3:2 for portraits, 21:9 for cinematic, 1:1 for social.
• --stylize 0–1000 Higher = more artistic, less faithful to prompt. Sweet spot: 250–500.
• --chaos 0–100 Variation across the four grid images. Use 15–30 to explore ideas.
• --weird 0–3000 Unusual aesthetics. Great for concept art.
• --sref [url] Style reference — copies aesthetic, not content.
• --cref [url] Character reference — keeps a character consistent across images.
• --no Negative prompt. e.g. --no text, watermark, extra fingers.
• --v 7 Explicit V7 model. Combine with --draft for fast iteration.

Some words punch far above their weight in V7. Layer these strategically:

• Photography: "shot on Kodak Portra 400", "Hasselblad", "cinematic still", "national geographic photography".
• Illustration: "gouache", "screenprint", "ligne claire", "Studio Ghibli style", "vector illustration".
• 3D / render: "octane render", "unreal engine 5", "PBR materials", "isometric 3D".
• Lighting: "golden hour", "chiaroscuro", "volumetric light", "neon rim lighting", "overcast diffuse".

1. "minimalist product photo of a matte black ceramic mug, seamless white studio backdrop, soft top light, macro detail on rim --ar 4:5 --stylize 300"
2. "cyberpunk street food vendor at night, neon signs reflecting in rain puddles, cinematic still, Blade Runner aesthetic, shot on Arri Alexa, 35mm --ar 21:9 --stylize 500"
3. "cozy attic reading nook, warm lamp light, rain on window, editorial interior photography, natural film grain --ar 3:2 --stylize 250"
4. "editorial portrait of a Nigerian fashion designer, colorful woven fabric background, natural window light, shot on medium format, sharp eyes --ar 4:5"
5. "isometric 3D scene of a tiny bakery cross-section, warm colors, soft studio light, octane render --ar 1:1 --stylize 400"
6. "gouache illustration of a paper airplane flying over mountains at sunrise, minimal palette, hand-painted texture --ar 16:9 --stylize 450"
7. "underwater documentary photo of a humpback whale mother and calf, sun rays piercing water, wide angle lens --ar 21:9 --stylize 250"
8. "1980s sci-fi book cover, chrome spaceship over alien desert, airbrush illustration, warm gradient sky --ar 2:3 --stylize 600"
9. "minimal architectural photo of a Tadao Ando concrete chapel interior, single ray of light, quiet mood, symmetrical composition --ar 4:5"
10. "street photograph of Tokyo alley in the rain at night, reflections, tungsten glow, shot on Leica Q3 --ar 3:2 --stylize 200"

• Long descriptive sentences instead of dense keywords — V7 penalizes rambling.
• Stacking too many contradictory styles ("photorealistic anime oil painting").
• Forgetting --ar and getting default square crops.
• Ignoring --stylize and wondering why images look "AI generic".
• Trying to prompt characters by name without --cref.

When a V7 image is close but not right, don't scrap the prompt. Change one variable at a time: swap the lighting term, bump --stylize, add a style reference, or use --vary (subtle) on the winning image. This isolates what actually moves the needle in your prompt.