V7 Parameters, Weights, and Personalization
Complete Parameter Reference
| Parameter | Range | Default | What It Does |
|---|---|---|---|
--chaos or --c |
0–100 | 0 | Controls variation between outputs. 0 = consistent, 100 = wildly different |
--stylize or --s |
0–1000 | 100 | How much Midjourney applies its own aesthetic. Low = literal, high = artistic interpretation |
--quality or --q |
.25, .5, 1, 2, 4 | 1 | Rendering quality/detail. Higher = more GPU time and detail |
--ar |
any ratio | 1:1 | Aspect ratio. Common: 16:9, 9:16, 3:2, 4:5, 21:9 |
--seed |
0–4294967295 | random | Reproducibility. Same seed + same prompt = similar result |
--no |
text | none | Negative prompt. Excludes concepts: --no text, watermark, blur |
--tile |
flag | off | Creates seamlessly tileable patterns |
--repeat or --r |
1–40 | 1 | Runs the prompt multiple times. Great for selection |
--personalize or --p |
flag or code | on (V7) | Applies your personalization profile to outputs |
--exp |
flag | off | Experimental features toggle |
--draft |
flag | off | Fast, lower-quality preview mode (V7 only). ~2x faster |
--oref |
image URL | none | Omni Reference — multi-purpose reference image |
--ow |
0–1000 | 100 | Omni Reference weight — how strongly to match the reference |
--sref |
image URL | none | Style Reference — match the style/aesthetic of a reference |
--sw |
0–1000 | 100 | Style Reference weight |
--stop |
10–100 | 100 | Stop generation early. Lower = softer, more impressionistic |
--iw |
0–3 | 1 | Image weight when using image prompts |
Prompt Weight Syntax
Midjourney uses the :: separator to assign weights to different parts of a prompt.
vibrant jungle scene:: colorful parrots::2 ancient ruins::0.5
- Default weight is 1 for any section without explicit weight.
colorful parrots::2— parrots get twice the emphasis.ancient ruins::0.5— ruins are de-emphasized, appearing as background elements.
More Examples
portrait photography::2 studio lighting:: dark background::1.5 shallow depth of field::
cyberpunk city::1 neon signs::2 rain::1.5 lonely figure::0.8
product photo:: glass bottle::2 water droplets::1.5 black background::1 --no label, text
Tip: Negative weights (e.g., trees::-0.5) can be used to actively suppress elements, similar to --no but with finer control.
V7 vs. V6: What Changed
V7 represents Midjourney's most significant upgrade. Key improvements:
| Area | Improvement |
|---|---|
| Anatomy | ~40% improvement in human anatomy accuracy. Fewer hand/finger errors, better body proportions |
| Prompt Understanding | ~35% improvement in following complex, multi-element prompts. Better spatial reasoning |
| Draft Mode | New --draft flag for rapid previewing at ~2x speed. Essential for iterative workflows |
| Personalization Default | Personalization is ON by default in V7. Your taste profile influences every generation |
| Omni Reference | New unified reference system replacing separate image/style/character references with a single, more intelligent --oref |
What V7 Still Struggles With
No model is perfect. Knowing the limitations saves you time:
- Text rendering — Still unreliable for more than 2-3 words. Use Gemini or post-production for text-heavy images.
- Hands and fingers — Better than V6 but not solved. Complex hand poses (holding small objects, interlaced fingers) still fail frequently.
- Character consistency — Improved with oref, but maintaining exact facial identity across many images remains challenging without external tools.
- Subtle expressions — Nuanced emotional expressions (wistful, sardonic, amused skepticism) are hard to control precisely.
- Complex scene coherence — Scenes with 4+ interacting characters or complex spatial relationships can break down.
The Personalization System
V7 introduced a three-tier personalization system:
1. Global Personalization
Your baseline taste profile, built from every image you've rated in Midjourney. Automatically applied to all V7 generations. Toggle off with --p 0 or --no personalization.
2. Standard Personalization
Community-shared style codes. Use with --p [code] to apply someone else's trained aesthetic.
3. Moodboard Profiles
Upload a curated set of images to define a custom personalization profile. The model learns the common aesthetic thread across your references.
Workflow:
- Rate 200+ images in Midjourney's ranking system to build your global profile.
- Test your profile: generate the same prompt with
--pon and off to see the difference. - For project-specific work, create moodboard profiles that override your global taste.
Exercise
Parameter Mastery Drill
- Write a single prompt (no parameters) and generate it. Save the result.
- Now generate the same prompt with each of these parameter combinations and note the differences:
--s 0vs.--s 500vs.--s 1000--c 0vs.--c 50vs.--c 100--q 1vs.--q 4--pvs.--p 0- With prompt weights emphasizing different elements
- Build a personal parameter cheat sheet: for each parameter, write one sentence about when you'd use a low value vs. a high value.