Midjourney: how to write prompts the model actually understands
MidJourney · Updated:
Midjourney is a family of image models from Midjourney Inc., available via midjourney.com and Discord. The lineup includes V7 (main), V8 Alpha (newest, web-only), Niji 6/7 (anime). All versions share the same parameter syntax (--ar, --s, --sref) and prompt hierarchy, but V7+ require natural language instead of tag lists.
What to know about Midjourney
Midjourney is a closed model with subscription pricing ($10/$30/$60/$120 per month). All versions return 4 variations per request, support any aspect ratio via `--ar`, and are optimized for English.
The key axis is version choice. V7 is the main one — best photorealism, natural language, broad availability. V8 Alpha is the literal-interpretation upgrade with native 2K via `--hd`, available only on alpha.midjourney.com. Niji 6/7 is the specialized anime model that understands manga and anime terminology natively. V5.2/V6 are considered legacy and not recommended without specific reason.
- Versions: V7 (main), V8 Alpha (web-only), Niji 6/7 (anime)
- Any aspect ratio via --ar
- Optimized for English
- 4 variations per request, up to 6000 characters in a prompt
- Negative prompt --no works reliably in V8
Prompt structure
Universal hierarchy for all versions: [Subject] + [Subject details] + [Context/setting] + [Style/mood] + [Camera/lighting] + [Parameters].
Opening words carry more weight — put the main subject in the first sentence. «Beautiful cinematic photo of a woman» (style first) underperforms «A confident businesswoman in her 40s, standing by floor-to-ceiling windows, soft daylight, 85mm lens».
For V7 and V8 write in coherent sentences, like a brief to a cinematographer. Comma-separated tag lists are an anti-pattern: «mountain, fog, sunrise, epic» loses to «A snow-capped mountain emerging from fog at sunrise, dramatic warm light on the peaks».
Core parameters
The base parameters are the same across versions: `--ar` (aspect ratio), `--v` (model version, 7 by default), `--s` or `--stylize` (0–1000, default 100 — interpretation strength), `--c` or `--chaos` (0–100, variation diversity), `--w` or `--weird` (0–3000, experimentation), `--no` (negative prompt), `--seed` for reproducibility.
Style controls: `--style raw` (minimal artistic treatment), `--p` (personalization), `--sref [URL]` (style reference) with `--sw` weight, `--iw` (image weight), `--niji` (switch to anime model). The `--cref` parameter (character reference) works only in Niji 6; it's removed in V7 and V8.
`--s` ranges: 0–50 for product/technical shots, 50–150 for general use (default), 150–300 for atmospheric work, 300–500 for illustration and concept art, 500–1000 for abstract and experimental.
Picking a version for the task
V7 is the default for most tasks: best photorealism, natural language, text rendering, personalization. Available on both Discord and web.
V8 Alpha is for tasks where you need native 2K via --hd, precise literal interpretation without artistic «enhancement», 5x faster generation than V7, or the most reliable negative prompt (--no works significantly better). Available only on alpha.midjourney.com — no Discord.
Niji 6/7 is for anime, manga, Eastern illustration. It understands terminology: «shoujo heroine», «cel shading», «mahou shoujo», «studio ghibli style». Niji 7 is the current recommended version; Niji 6 is only needed if you require style presets (--style expressive/cute/scenic) or --cref.
Common mistakes
1. Comma-separated keyword lists
The headline V7+ anti-pattern. «cat, fluffy, cute, big eyes, garden, flowers, sunny» is worse than «A fluffy cat with big curious eyes, sitting among wildflowers in a sunny garden». Pre-V6 versions tolerated commas; V7 and V8 expect coherent grammar.
2. Quality-spam at the start of the prompt
«beautiful, stunning, 8k, masterpiece, best quality» — words with no visual information that burn positional weight at the front. V7 and V8 ignore them entirely. Replace with concrete anchors: lens, light type, genre.
3. Style before subject
«Beautiful cinematic photo of a woman» — style is first, subject is third. The model weights opening words most heavily, and style steals the priority. Move «who or what» to the front.
4. Using legacy versions without reason
V5/V6 are significantly weaker than V7 in anatomy, photorealism, and prompt comprehension. Unless you have a specific reason (e.g. --cref in Niji 6), use V7 or V8. Old versions are technical debt.
5. Changing --ar mid-iteration
Aspect ratio drives composition. Switch --ar from 1:1 to 16:9 and you're effectively starting over because composition rebuilds entirely. Decide --ar at the start of the project and don't change it mid-iteration.
Before / after examples
Example 1
Before
beautiful, stunning, 8k, detailed, mountain, fog, sunrise, epic, cinematic, masterpiece
After
A snow-capped mountain emerging from morning fog at sunrise, dramatic warm light catching the peaks, wide-angle composition, national geographic photography, atmospheric haze --ar 16:9 --s 150 --v 7
Quality-spam plus a tag list — a double anti-pattern. A coherent description with specific light and a genre anchor («national geographic photography») reliably outperforms.
Example 2
Before
anime girl with sword
After
A fierce warrior with wild red hair and golden eyes, mid-leap wielding a flaming katana, slashing through dark energy, speed lines and dynamic motion blur, dramatic backlighting, shonen anime style --ar 16:9 --s 300 --niji 7
For anime, switch to Niji via --niji 7. Niji understands anime-specific vocabulary (shonen, speed lines, dynamic motion blur, backlighting) natively.
Example 3
Before
product photograph of watch
After
Product photograph of a luxury watch on black marble surface, single overhead softbox light, clean white background, sharp focus, commercial photography --ar 1:1 --s 25 --style raw --v 7
Product photography needs low `--s` (25) plus `--style raw` — this removes the artistic interpretation that breaks commercial usage.
