Midjourney v7 Guide: --sref, --cref, Draft Mode & New Parameters (2026)

Midjourney v7 launched in early 2026 with a set of new parameters that have fundamentally changed how creators work with the model. If you're still using v6.1 workflows, you're leaving significant quality and consistency on the table.

Here's everything you need to know about v7's most important additions.

What's new in Midjourney v7

v7 introduced four major changes from v6.1:

• --sref (style reference) — now more powerful and consistent
• --cref (character reference) — maintain a character across images
• Draft Mode — fast, cheap generation for iteration
• Improved prompt understanding — natural language works better; less need for keyword stacking

--sref: Style Reference

--sref [image URL] applies the visual style of a reference image to your prompt. v7 made --sref dramatically more stable — the style transfers more faithfully without bleeding into composition.

Basic syntax: [your prompt] --sref [image URL]

With style weight: [your prompt] --sref [image URL] --sw 50

Style weight (--sw) controls how strongly the style is applied. Range: 0–1000. Default: 100. Higher = stronger style transfer. Lower = more freedom from the style.

Practical example: Want all images in a project to share a consistent muted film photography aesthetic? Generate one reference image, then use its URL as --sref across every generation in that project.

Multiple style references: --sref [url1] [url2] — blends both styles. Weight them individually: --sref [url1]::2 [url2]::1

--cref: Character Reference

--cref [image URL] maintains a consistent character appearance across different scenes, poses, and environments. This is the feature that creators have been requesting for years.

Basic syntax: [description of scene with character] --cref [character image URL]

Character weight: --cw [0–100] — controls how closely the character appearance is followed. 100 = maximum fidelity (face, hair, body). 0 = only general style preserved.

Best practices for --cref:

• Use a clean, well-lit reference photo with the character facing forward
• Use --cw 75–85 for natural variation across scenes while keeping the character recognizable
• Combine with --sref for both consistent character AND consistent visual style
• Works best with real human photos; results vary more with illustrated characters

Workflow example: A woman walking through a busy Tokyo intersection, night, neon lights reflecting on wet pavement --cref [character URL] --cw 80 --sref [style URL] --sw 80 --ar 2:3

Draft Mode

Draft Mode generates images at roughly 4× the speed of standard v7 — at lower quality. It's designed for rapid iteration before committing to a high-quality render.

Enable: --draft parameter or toggle in the web interface

Use Draft Mode for: exploring composition options, testing prompt variations, narrowing down concepts before final render

Typical workflow: Generate 8–12 Draft Mode variants → pick the strongest 2–3 → render those at full quality

Draft Mode costs approximately 25% of standard generation credits. For high-volume creators, this significantly extends your subscription's effective capacity.

Updated --stylize behavior in v7

v7 recalibrated the --stylize range. Behaviors:

• --stylize 0 — strict literal prompt following, minimal aesthetic enhancement
• --stylize 100 — default, balanced
• --stylize 400 — stronger Midjourney aesthetic interpretation
• --stylize 1000 — maximum artistic freedom, loosest prompt adherence

v7 with --stylize 200–400 produces noticeably more cinematic, polished results than the equivalent in v6.1.

The v7 prompt structure

v7's improved language understanding means you can write more like a director and less like a search engine:

v6.1 style (still works): woman, forest, golden hour, shallow depth of field, film grain, Leica --ar 4:5 --style raw

v7 optimized style: A woman stands at the edge of a pine forest at golden hour, warm light filtering through the canopy. Shot on 35mm film with a Leica, shallow depth of field, subtle grain. --ar 4:5 --stylize 200

Both work — but v7 responds better to the second format, especially for complex or nuanced scenes.

10 v7-specific prompt examples

• Bird's eye view of a river delta at sunset, fractal water patterns, aerial photography, muted earth tones --ar 16:9 --stylize 300
• An elderly monk reading in a lantern-lit temple corridor, dust particles in the beam of light, 35mm film photography, --style raw --ar 3:2 --stylize 150
• Macro photograph of frost crystals forming on a dark window pane, clinical sharpness, blue-black tones, microscopic detail --ar 1:1
• A woman in a 1970s interior, warm tungsten light, sitting at a cluttered desk, candid, documentary style --cref [url] --cw 80 --ar 4:5
• Abstract architectural photography of a modern staircase, minimal, black and white, geometric shadows, negative space --ar 2:3 --stylize 400

Using a prompt generator for v7

Keeping up with each new Midjourney version's parameter changes is time-consuming. A Midjourney-aware prompt generator stays current with v7 syntax and outputs correctly formatted prompts — including --sref, --cref, and --stylize recommendations — from your plain language description.