Midjourney V7 Prompt Guide 2026: From Beginner to Commercial-Grade Image Generation

Midjourney produces stunning images, but there's a gap between "a casual sentence" and "commercially deliverable." This article provides: a standard prompt structure, V7 core parameter quick reference, recipe library by use case, and the most critical consistency control for commercial delivery.

1. Standard Prompt Structure

Organize by the five-part "Subject → Scene → Lighting → Style → Parameters" structure, with one or two phrases per part:

text
a weathered fisherman mending nets,      ← Subject (specific action is better than static description)
on a misty harbor dock at dawn,           ← Scene
soft golden backlight, fog diffusion,     ← Lighting (biggest variable for image quality)
documentary photography, 35mm film grain, ← Style
--ar 3:2 --stylize 200                    ← Parameters

Key points: The weight of lighting words is widely underestimated — golden hour / rim light / soft window light / neon glow; changing lighting changes the image more than changing style words. Use --no for negative content (e.g., --no text, watermark) instead of writing "no text."

2. V7 Core Parameter Quick Reference

ParameterFunctionCommon Values

--arAspect ratio16:9 landscape / 3:4 portrait / 1:1 avatar --stylize (--s)Aesthetic processing strength0-100 faithful reproduction, 250-750 artistic flair --chaosVariation among four images0 stable, 30+ for inspiration --weirdUnconventional aestheticsFor exploratory output --seedFix random seedFor controlled variations with prompt tweaks --cref + --cwCharacter consistency reference (V7 also has omni-reference usage)--cw 100 strictly maintains appearance --sref + --swStyle referenceTransfer style from an image to new content --draft / Fast modeLow-cost draft iterationDraft first, refine later to save credits

3. Recipe Library by Use Case

E-commerce Product Image:

text
[product] product photography, on minimal stone pedestal, studio softbox lighting,
clean beige background, commercial shot, ultra sharp --ar 1:1 --s 100

WeChat Official Account / Blog Header Image:

text
flat illustration of [theme concept], geometric shapes, limited palette of
deep blue and coral, generous negative space for text --ar 16:9 --s 250

Portrait / Headshot:

text
portrait of [character description], soft window light from left, shallow depth of field,
85mm lens look, natural skin texture --ar 3:4 --s 150

Game / Novel Concept Scene:

text
ancient mountain monastery above sea of clouds, epic scale, volumetric
god rays, matte painting style, artstation quality --ar 21:9 --s 400

Each recipe is a skeleton — replace the subject, keep the lighting and parameter parts. This is the fastest way for beginners to consistently produce good images.

4. Key to Commercial Delivery: Consistency

Clients don't want just one good image; they want a set of images with the same character/style:

Character Consistency: First generate a satisfactory character portrait → then attach --cref portraitURL --cw 100 to all subsequent images; lower --cw if you need different expressions/poses.

Style Consistency: Attach the same --sref to the entire set, or fix a "lighting + style + parameters" suffix template.

Reproducibility: Record the full prompt and seed for each image in the delivery document — a lifesaver when the client wants changes two weeks later.

Batch Variations: Use --repeat with the same prompt for multiple sets, or fix the seed and tweak wording — more controllable than random regeneration.

For pricing and copyright boundaries (AI-generated copyright registration issues, platform commercial terms), see AI Illustration Monetization Guide.

5. Beyond Midjourney

For precise control (pose/composition/ControlNet) or zero marginal cost batch generation → local Stable Diffusion 3.5

Trade-offs among three major image models → Midjourney vs DALL·E vs SD

Prompt philosophy is universal: structured description + parameter control works on any model (underlying principle: Diffusion Models and CFG)

FAQ

Q: Are longer prompts better? No. V7 has strong natural language understanding; 20-40 informative words is better than piling up 100 tags; ask yourself for each added word "what does it change in the image?"

Q: Can I use Chinese prompts? Yes, but English has richer training data; professional photography/art terms have significantly higher hit rates in English — Chinese ideas + English keywords is a practical compromise.

Q: What's the fastest way to practice? Learn from others: find images you like in the community, view the full prompt, replace the subject while keeping the lighting and style parts — reverse-engineering 20 images beats blindly trying 200.

---

*Last updated: June 2026. Version features and pricing are subject to Midjourney's official announcements.*