Midjourney V7 Complete Guide 2026: Full Parameter Breakdown + High-Quality Image Tips

From beginner to consistent output, systematically master the core usage of MJ V7

Midjourney V7 is one of the highest commercial quality AI image generation models available. Compared to V6, V7's prompt understanding has made a qualitative leap—effects that previously required complex prompts can now be achieved with simple descriptions.

But "better understanding" also means—if your prompting methods are still stuck in the V5/V6 era, your results on V7 might actually get worse.

Key Changes Between V7 and V6

Prompt Understanding

V6: Requires explicit style words (e.g., "cinematic lighting, 4k, hyperdetailed")

V7: Understands natural language descriptions; too many technical terms can interfere

Character Consistency

V6: Hard to keep the same character consistent across different images

V7: The --cref parameter greatly improves character consistency

Speed

V7 Turbo mode is about 4x faster than V6

How to Write V7 Prompts Correctly

Basic Structure


[subject description] [scene/environment] [lighting/atmosphere] [style]

V6 Era (Verbose Style):


portrait of a young woman, cinematic lighting, bokeh background, 8k resolution, hyperdetailed, photorealistic, shot on Canon EOS R5, f/1.4 aperture, golden hour

V7 Recommended Style (More Natural):


portrait of a young woman in golden hour light, soft bokeh, photographic, high quality

V7 understands that "golden hour" implies warm tones, side lighting, and soft shadows—no need to list technical parameters one by one.

Chinese Prompts

V7 supports Chinese prompts, but a mix of Chinese and English is recommended:


一位穿汉服的年轻女性，站在古典园林中，soft natural light, photorealistic

Pure Chinese works too, but English image vocabulary (like light, bokeh, cinematic) is still more precise than Chinese.

Key Parameters Explained

--ar (Aspect Ratio)

The most commonly used parameter:


--ar 16:9   # Horizontal video cover
--ar 9:16   # Vertical phone wallpaper / Xiaohongshu image
--ar 1:1    # Square (avatar, social media post)
--ar 4:3    # Traditional photo ratio
--ar 3:2    # Common photographer ratio

--style (Style Strength)


--style raw      # Stays closer to the prompt, reduces MJ's "beautification"
--style cute     # New in V7, cute style

For commercial images requiring precise control, usually add --style raw.

--cref (Character Reference) — Important New V7 Feature


/imagine portrait of [character name] --cref [previous image URL]

This feature lets you upload a character image and generate images of that character in different scenes/poses while maintaining consistent appearance. Very useful for comic creation and brand character design.

--sref (Style Reference)


/imagine [description] --sref [reference image URL]

Makes the generated image's style match the reference image (color tone, brushwork, artistic style, etc.) without copying the content.

--chaos (Creativity Randomness)


--chaos 0    # Most stable, close to prompt description
--chaos 50   # Balance between creativity and stability
--chaos 100  # Maximum randomness, wildly imaginative

Use high chaos when exploring styles, then dial back to 0 for stable output once direction is set.

Advanced Techniques

4.1 Batch Testing Prompt Variations


/imagine {minimalist, maximalist, brutalist} poster design for tech conference, [description]

Use curly braces to generate multiple variations at once, helping quickly find the best direction.

4.2 Weight Control


red apple::2 on a wooden table::1

The number after :: is the weight; higher values make that element more prominent.

4.3 Negative Prompts


beautiful landscape --no people, buildings, cars

--no tells MJ to avoid certain elements.

Commercial Usage Notes

Pro subscription and above have commercial usage rights

Generated images are publicly displayed in the MJ community by default; sensitive projects require Stealth Mode (Pro feature)

Generated character images cannot be used for deceptive purposes (impersonating real people, etc.)

Check text in generated images—MJ V7's text rendering has improved but can still make mistakes