Midjourney essentials: prompts, parameters, style references

Midjourney is a tool with a dense, undocumented-feeling vocabulary. Once you learn it, your hit rate goes from 10% to 70%.

The prompt structure that works

A working Midjourney prompt has shape:

[subject], [action/context], [environment], [style], [lighting], [camera/composition] --parameters

Example:

A lighthouse, standing resolute against a stormy sea, rocky coastline,
oil painting in Turner's style, dramatic golden-hour light, wide shot
from below --ar 16:9 --style raw

Missing a section gets you generic output. Filling them all gets you specific output.

Key parameters

• --ar 16:9 aspect ratio. 9:16 for vertical, 1:1 for square, etc.
• --style raw less Midjourney-aesthetic-default, more literal.
• --stylize 100 (default) / --s 0–1000. Lower = closer to prompt; higher = more artistic.
• --chaos 50 controls variety. 0 = tight set of similar results; 100 = wild variations.
• --v 6 model version. Newer usually better; some styles reserve certain versions.
• --no hands negative prompting. Remove elements you don't want.

Style references (--sref)

The biggest Midjourney upgrade of the last year: style references let you anchor output to a specific image's aesthetic:

lighthouse in a storm --sref https://example.com/painting.jpg

Now your output takes styling cues from that image. Use for consistent series, brand alignment, or matching a reference mood.

Character references (--cref)

Keep a character consistent across multiple generations:

[subject as defined by the reference] doing something --cref https://example.com/character.jpg

Still imperfect but much better than without.

Iteration strategy

• First batch: cast a wide net (chaos 40-60, stylize default).
• Identify what works and what doesn't.
• Next batches: narrow with more specific prompts, lower chaos, style references.
• Upscale the best candidate.

Don't expect the first generation to be the final. 5-10 iterations is normal.

The vocabulary that matters

Words that shift output significantly:

• Camera terms: "35mm film," "wide-angle," "shallow depth of field," "tilt-shift."
• Lighting: "golden hour," "rim light," "chiaroscuro," "bioluminescent."
• Style: "oil painting," "woodcut," "anime," "photorealistic."
• Mood: "melancholy," "frenetic," "serene," "menacing."

Build a personal prompt vocabulary. Most pros end up with a notes doc of prompts that work.

What Midjourney is bad at

• Text in images. Use Ideogram for that.
• Consistent faces across many images. Character reference helps; still imperfect.
• Complex scenes with many specific elements. Gets confused past ~3 subjects.
• Precise object counts. "Seven apples" often gives six or eight.
• Anatomy of hands. Famously troublesome; improving but still a risk area.

The commercial use question

Midjourney's Pro tier grants commercial use rights to outputs you generate. Verify current terms for your tier — they evolve.

For client work, document the generation process (prompts, settings, reference images). Protects you if rights are later questioned.