Midjourney v7 Prompt-Guide — Was 2026 wirklich funktioniert

11 minute read · Published April 2026

Midjourney v7 (released late 2025) changed how prompts should be written. Many guides online still reference v5 or v6 patterns — stuffing "8k ultrarealistic masterpiece" tags, relying on --no parameters, treating one prompt as a one-shot goal. None of that is current best practice.

This guide covers what actually works in v6.1 (stable) and v7 (current), calibrated to the 2026 reference-mode workflow that separates professional output from lucky accidents.

The Three 2026 Shifts You Need to Know

1. Reference modes (--sref, --cref) replaced "hope for consistency." Before v6, generating 5 cohesive images required running same prompt 50 times. Now: one strong anchor image + --sref on follow-ups = tight consistency across series. Game-changer.

2. --style raw for photorealistic is non-negotiable. Default Midjourney adds subtle stylization. Without --style raw, your "professional photograph" prompt produces something subtly stylized. --style raw is the fix.

3. Personalization (--p) in v7. After 200+ ratings, your aesthetic becomes a parameter. Most under-used v7 feature.

Before building prompts, understand these three shifts. The rest of this guide assumes you're working within them.

The Correct Prompt Structure

2026 prompt order matters more than it did in earlier versions:

Subject → Style → Setting → Lighting → Camera/Technical → Parameters

Earlier words weight heavier. So:

```

✓ Minimalist Scandinavian workspace, soft morning sunlight through linen curtains, wooden desk with ceramic coffee mug, warm muted palette, shallow depth of field, editorial photographic style, analog film aesthetic --ar 16:9 --style raw --stylize 300 --v 7

```

NOT:

```

✗ --ar 16:9 --style raw --stylize 300 --v 7, ultrarealistic masterpiece 8k, desk with mug, Scandinavian

```

This is the structured approach we use in the Midjourney Mastery Pack — one of our most-used prompt patterns across AI artists.

Parameter Reference (What Actually Matters)

--style raw (critical for photorealistic)

Without --style raw, your outputs have subtle stylization. With it, outputs are more literal to your description. For product photography, editorial photography, any photorealistic intent — always include.

--stylize (value 0-1000, default 100)

--stylize 50: Very literal to prompt. Use when you want Midjourney to NOT interpret.
--stylize 250-500: Sweet spot for most work. Balanced prompt-literalness + artistic quality.
--stylize 800-1000: Midjourney has more artistic liberty. Use for creative/painterly work.

Rarely use default 100. Tune deliberately.

--chaos (value 0-100, default 0)

--chaos 0: Default. Four images in grid are similar.
--chaos 50: Moderate variation. Useful for ideation phase.
--chaos 100: Wildly different interpretations. Brainstorming territory, not production.

Use for ideation. Never for production (you want consistency).

--weird (value 0-3000, v6.1+ feature)

--weird 500: Controlled creative surprises. Unusual compositions, unexpected choices.
--weird 1500-3000: Genuinely weird territory. Good for concept art exploration; not for clients who want predictability.

--sref [URL] --sw [0-1000] (style reference, v6.1+)

The 2026 game-changer. Paste a previous Midjourney image URL; --sref tells Midjourney "match this aesthetic."

--sw 100 (default): Strong style inheritance + content flexibility.
--sw 50: Balanced. More content freedom; looser style match.
--sw 200+: Near-replicate style (limits content flexibility).

For series/books/brand work: use --sref on all images after building one strong anchor. See the Style Anchor Builder for the full workflow.

--cref [URL] --cw [0-100] (character reference)

Same but for character consistency. Essential for children's books, graphic novels, brand mascots. Start --cw 80-100 for strict consistency; relax to 50 if outputs feel rigid.

See Midjourney Character Consistency for the multi-angle anchor library workflow.

--p (personalization, v7 only)

After 200+ ratings (intentional, not random swipes), Midjourney learns your aesthetic. Adding --p to prompts pulls from your rating history.

Massively under-used feature. If you rate intentionally over 2-4 weeks, --p produces aesthetically-aligned output that strangers' prompts can't replicate.

Aspect Ratios (--ar)

Match your target platform:

--ar 16:9: Video thumbnails, blog heroes, desktop headers
--ar 9:16: Vertical phone social (TikTok, Instagram Stories, Reels)
--ar 1:1: Instagram feed, podcast cover, general square
--ar 2:3: Print portrait, book cover
--ar 3:2: Print landscape, photographic standard

Check target first; choose aspect ratio; commit in prompt.

Negative Prompts (--no) Are Weak in Midjourney

Unlike Stable Diffusion where negative prompts are critical, Midjourney's --no parameter is weak. Most users waste tokens on --no lists that don't affect output much.

Better: describe what you WANT, not what you don't want.

Weak: cleanup kitchen scene --no clutter, --no dirty dishes, --no mess
Strong: minimalist kitchen scene, single clean surface, empty counters, organized, pristine

Positive descriptions guide better than negative exclusions.

Category 1: Portraits + People

Key insight: Midjourney is best at stylized portraits, less reliable for photorealistic human likeness.

For stylized portraits:

```

Portrait of [character], [age], [distinctive features], [style reference — watercolor / editorial illustration / painterly / anime / etc.], [composition — close-up / 3-quarter / full body], [lighting — golden hour / studio softbox / chiaroscuro], --ar [preferred] --style raw OR [stylized parameter as appropriate] --v 7

```

For photorealistic: budget extra iterations. Subtle "AI look" (plastic skin, weird hands, mirror-symmetric faces) is harder to eliminate in photorealistic portraits than stylized ones.

Category 2: Product Photography

Studio-quality product shots require structured prompting. Generic "product photo of X" produces cartoon-ish results.

The winning structure:

```

[Product] with [specific material], [specific size/scale context], [background — white seamless / warm wood / concrete], [lighting direction — 45-degree softbox / overhead diffused / rim light], [depth of field], [register — editorial / commercial / catalog], [camera spec if relevant — 85mm lens, f/2.8] --ar 1:1 --style raw --stylize 150 --v 7

```

See the Product Photography Prompt Builder for a complete e-commerce workflow.

For Amazon main image requirements (pure white RGB 255): Midjourney produces close but rarely exact. Post-process through Photoroom or remove.bg to hit technical spec.

Category 3: Editorial Illustration

Editorial illustration has specific conventions: limited palettes (3-5 colors), bold simplified shapes, conceptual rather than literal composition.

Don't: Try for photorealistic editorial ("editorial illustration of person reading book")

Do: Build for limited-palette flat-shape aesthetic ("editorial illustration with limited ochre cream and slate palette, flat bold shapes, conceptual composition with negative space, mid-century editorial register")

See Midjourney Editorial Illustration Framework for the conceptual-vs-literal distinction that separates amateur from professional editorial work.

Category 4: Series + Book Work

For 5-30+ image projects (children's books, brand campaigns, concept art portfolios):

Spend 20-30% of project time on ideation. 48+ images across 3 direction explorations. Don't shortcut.

Select anchor deliberately. Not "favorite" — "most productionable direction."

Use --sref on all follow-ups. Text alone cannot maintain consistency across generations.

--sref + --cref stack for character + style consistency simultaneously.

The full workflow is documented in Midjourney Ideation-to-Production Pipeline.

Common Mistakes That Still Happen

Keyword stuffing: "8k ultrarealistic masterpiece award-winning photography cinematic professional" — Midjourney v6.1+ treats these as dilution, not enhancement. Use specific visual descriptors instead.

Relying on --no: Describe what you want. Negative prompts are weak in Midjourney.

No aspect ratio specification: Default is 1:1. Without --ar, you get square. Check target platform first.

Mixing --style raw + stylization intent: --style raw and "whimsical" fight each other. Choose one register per prompt.

Forgetting --v 7 parameter: Without explicit --v 7, sometimes outputs use older version behavior. Be explicit.

Where to Go From Here

If you're starting Midjourney:

Learn the parameter system first (Category 2 prompts in the Mastery Pack)
Spend first 30 days on ONE aesthetic (not exploring everything)
Rate 200+ images intentionally to unlock --p personalization

If you're intermediate:

Master --sref workflow (Style Anchor Builder)
Learn editorial illustration vs. photorealistic distinction
Stop relying on --no; describe positive

If you're advanced:

Build anchor libraries for different project types
Document your own aesthetic parameters
Teach others (or use Promptolis prompts with team)

FAQ

Q: Do I need Midjourney Pro for commercial use?

Yes. Check current Midjourney terms; paid plans grant commercial license, free-tier doesn't.

Q: Is Midjourney better than Stable Diffusion / Flux for my use case?

Depends. Midjourney = easiest + most aesthetic-coherent. Stable Diffusion = most customizable + on-prem option. Flux = newer, strong for photorealistic + text-in-image.

Q: Can Midjourney replace real photography for product shots?

For simple products + minimal budget: often yes. For luxury / brand-critical / reflective materials: hybrid workflow (AI + real photography + post-processing) usually better.

Q: What's the future of Midjourney (v8 etc.)?

Midjourney's pace is unpredictable. Core patterns (reference modes, structured prompts, --style raw) should remain relevant. Specific parameter values may shift.

Q: Does this guide apply to Midjourney on Discord vs. web?

Yes. Both interfaces use same underlying model. Web has better organization; Discord has community + conversation.

---

This article is part of the Promptolis library — free, research-backed AI prompt resources at promptolis.com.