Midjourney v6 vs v7 Prompts: What Changed and What Works
Why Your Old Prompts Suddenly Look Different
If you upgraded to Midjourney v7 and your carefully tuned v6 prompts started producing softer, more photographic, and sometimes less “on-style” images, you are not imagining it. v7 changed how the model reads prompts: it leans harder on natural-language understanding and treats long keyword stacks differently than v6 did. Prompts written for v6 are not broken, but several habits that v6 rewarded now work against you. This is a perfect Structure B case: old v6 style vs new v7 style.
For the fundamentals of building a prompt from scratch, our Midjourney prompt guide covers the base anatomy that still applies.
Wrong Prompt: The v6 Keyword Soup
This is the classic v6 habit: a dense pile of comma-separated modifiers, weighted heavily, with little sentence structure.
/imagine prompt: cinematic portrait of a warrior woman, highly detailed, 8k, unreal engine 5, octane render, dramatic lighting, bokeh, sharp focus, trending on artstation, masterpiece, --ar 2:3 --v 6.0 --s 750What goes wrong in v7: the model now parses meaning more like a language model and less like a tag matcher. Stuffing redundant quality tags (“8k,” “masterpiece,” “trending on artstation”) adds noise instead of signal. Heavy –s (stylize) values that looked great in v6 can flatten v7’s improved realism into over-designed mush. The prompt also ignores v7’s better instruction following.
Right Prompt: Natural Language With Intent
The fix is to write like you are describing the shot to a person, lead with the subject and the feeling, and let v7’s language understanding do the work. Reserve parameters for the few things that truly need them.
/imagine prompt: A weathered warrior woman stands at the edge of a cliff at dawn, wind pulling at her cloak, soft golden light catching the edge of her sword. Photographic, intimate, shallow depth of field, quiet tension. --ar 2:3 --style raw --v 7Notice: one clear sentence, a mood word (“quiet tension”), and only two parameters. –style raw keeps v7 from over-smoothing; we dropped the stylize weight entirely because v7 defaults are stronger.
Side-by-Side Comparison
| Aspect | v6-style prompt | v7-style prompt |
|---|---|---|
| Structure | Keyword stack | Natural sentence |
| Quality tags | Many (“8k”, “masterpiece”) | Almost none |
| Stylize | High (–s 750) | Default or low |
| Subject lead | Buried in tags | First and clear |
| Style flag | Often omitted | –style raw for realism |
| Version | –v 6.0 | –v 7 |
Why the Right Version Works
v7’s training shifted toward understanding prompts as language, so a coherent description gives it a stronger mental image than a bag of tags. Redundant quality words in v6 acted as weak emphasis; in v7 they compete with the real subject for attention. Lower stylize lets v7’s native photorealism show instead of forcing a v6-era “render” look. The –style raw flag is the single most useful new lever for people who want photographic results rather than the smoothed default aesthetic.
If you generate product shots, the same natural-language shift helps our product mockup prompt land more convincingly on v7.
How to Migrate Your Own Prompts
Run this conversion pass on any v6 prompt:
- Rewrite the modifier list as one or two descriptive sentences.
- Delete redundant quality tags; keep at most one mood word.
- Remove or lower –s unless you specifically want stylization.
- Add –style raw when you want photographic, less smoothed output.
- Swap –v 6.0 to –v 7, then re-roll and compare.
A migration template:
Old (v6): [subject], [tag], [tag], [tag], --ar X --v 6.0 --s 750
New (v7): [One sentence describing the subject, setting, light, and mood]. --ar X --style raw --v 7Keep aspect ratio; drop the keyword soup. Test one old prompt side by side with its v7 rewrite before migrating the whole library.
v7 Parameter Cheat Sheet
These are the flags worth knowing when moving from v6 to v7. Defaults changed, so what you set matters more than before.
| Flag | v6 habit | v7 recommendation |
|---|---|---|
| –v | 6.0 | 7 |
| –s (stylize) | 250-750 | 0-250, or omit |
| –style raw | rarely used | default for realism |
| –ar | keep | keep |
| quality tags | stacked | dropped |
The biggest mental shift is trusting the model to understand a sentence. You no longer need to spell every modifier as a tag. Write the shot the way you would describe it to a photographer, and v7 will usually surprise you with how much it caught.
When to Keep v6 Habits
Not everything from v6 should go. If you deliberately want a stylized, illustrative look rather than photorealism, a moderate –s still helps, and v6 remains available via –v 6.0 for legacy projects where you tuned a specific aesthetic. The rule: keep v6 style when the goal is stylization; switch to natural language when the goal is realism or you are starting fresh. Test both on one image before committing a whole batch.
Three v7 Prompt Starters by Genre
Use these as openings, then add your subject. They follow the natural-language rule and lean on –style raw where realism helps.
Photographic portrait:
A candid portrait of [subject], shot on 50mm, window light, skin texture visible, calm expression. --ar 4:5 --style raw --v 7Product on clean background:
A [product] centered on a seamless white background, soft studio light, faint shadow, e-commerce look. --ar 1:1 --style raw --v 7Illustration with stylization:
An editorial illustration of [scene], bold flat colors, textured paper feel, confident line work. --ar 3:2 --s 200 --v 7Notice the illustration one allows a little –s because the goal is stylization, exactly the keep-v6-habits exception from above.
Sources & Notes
Behavior described reflects Midjourney v7 as released in 2025 and contrasts with v6. Parameter names (–ar, –s, –style raw, –v) follow Midjourney’s documented syntax; always confirm current flags in your account since the model and flags evolve. Examples are illustrative; render results vary per seed. No external APIs or WordPress calls were made to produce this article.