Midjourney advanced: sref, --niji, blends, and the editor

Past the essentials, Midjourney rewards real fluency — style and character references, multi-prompt blending, the editor, niji, and the discipline of iteration.

Style reference sets

Single style references are powerful. Sets are more powerful.

Build a library:

• --sref <url1> <url2> <url3> blends styles.
• --sw 500 adjusts style weight (how strong the reference pull is).

Use cases:

• Brand coherence. Your brand's Midjourney style = a specific sref set.
• Series continuity. Multiple images that feel related without being identical.
• Genre blending. "Noir + Edward Hopper + contemporary photography."

Random style codes (e.g., --sref 1234567890) generate unique aesthetics; save the ones that work.

The editor

The Midjourney editor (web interface) lets you:

• Inpaint — select a region, regenerate just that region with a new prompt.
• Pan — extend the image in a direction, filling in new content.
• Zoom out — generate a larger canvas with the original as a subset.
• Vary region — selectively vary parts of an image.

These turn one-shot generation into iterative refinement. Most professional use is editor-heavy.

Niji mode

--niji produces anime / illustration style. Genuinely distinct aesthetic; worth its own tutorial.

• --niji 6 is the current generation; richer than older versions.
• Pair with style references for specific manga or anime aesthetics.
• Strong on character designs and expressive scenes.
• Less strong on realism (obviously) and Western comic styles.

Blending images

/blend merges 2-5 images into one. Works for:

• Combining a character and a setting.
• Mood-boarding several references into a unified composition.
• Generating variations on a theme.

Blend is less controllable than prompt; works best when the source images already share properties.

Multi-prompt weighting

:: splits a prompt:

mountain cabin :: snowstorm :: cinematic lighting::

Number after each :: sets relative weight:

mountain cabin ::2 snowstorm ::1 cinematic lighting ::1

Cabin dominates. Useful when the main subject is getting drowned out by atmospheric elements.

Consistency at scale

For a series (e.g., a marketing campaign, a book of illustrations):

1. Develop a style reference set early.
2. Define character references for recurring subjects.
3. Establish a prompt template; vary only the specific content per image.
4. Document the winning parameters.

Consistency is achievable; it requires discipline upfront.

Iterating with upscales

Midjourney's upscale options — Subtle, Creative, Custom Zoom — vary quality. Strategy:

• Subtle upscale for final output — preserves composition.
• Creative for adding detail / texture.
• Custom zoom for reframing.

Upscale only the top 1-2 candidates per session. Upscaling everything wastes compute and doesn't improve selection.

The prompt library discipline

Midjourney pros keep structured records:

• Prompt | Parameters | Sref | Output | Notes.
• Tagged by project, style, outcome.
• Shared with collaborators.

This isn't optional for serious production. Memory isn't good enough; prompts that worked last month fade.

Failure modes

• Prompt overload. 40-word prompts produce muddled output. Cut back to essentials.
• Style reference drift. Too many references cancel each other out. 1-3 is the sweet spot.
• Stylize too high on literal subjects. Medical diagrams with stylize 1000 are not medical diagrams.
• Over-iterating. If 20 tries haven't produced a viable result, the prompt structure is wrong — restart, don't tweak.