Midjourney v6 portrait prompts: a practical guide to more consistent faces

Midjourney responds best to short, clear descriptions. Write them in this order:

1. Subject: age range, hair, expression, clothing.
2. Framing: close-up, head and shoulders, half body.
3. Lens and camera: 85mm portrait lens, shallow depth of field.
4. Lighting: soft window light, golden hour, studio softbox, rim light.
5. Setting and mood: plain studio backdrop, rooftop at dusk, warm and calm.
6. Settings: aspect ratio and style options at the end.

close-up portrait of a woman in her late 20s, long dark wavy hair, gentle smile, ivory silk blouse, 85mm lens, shallow depth of field, soft window light from the left, neutral grey studio backdrop, editorial photography --ar 4:5 --style raw --v 6

• --ar 4:5 or --ar 2:3: portrait-shaped images that suit Instagram and print.
• --style raw: less of Midjourney's default look, so results feel more like real photos.
• --stylize (--s): lower values follow your words more closely; higher values add more artistic flair.
• --no: leave out things you don't want, for example --no text, watermark.

Midjourney v6 added a character reference option. Add --cref followed by an image link to ask Midjourney to keep the person's look. --cw (character weight, 0 to 100) controls how much it copies: 100 keeps face, hair and clothing; lower values keep mainly the face so you can change outfits.

the same woman walking through a flower market, white linen dress, morning sunlight, 50mm lens --ar 4:5 --style raw --v 6 --cref [your image link] --cw 40

Be realistic about what this does:

• It works best with characters that Midjourney itself created. Real photos of people are matched less closely.
• Small details such as freckles, scars or tattoos often change between images.
• Newer Midjourney versions use a different reference option (Omni Reference). Check Midjourney's own documentation for the version you use.

• Reuse the same description. Keep one fixed sentence for the person (hair, age, features) and only change outfit, pose and scene.
• Change one thing at a time. Changing lighting, lens and setting all at once makes faces drift more.
• Keep the same lens and framing. A 24mm close-up changes face shape compared with an 85mm shot.
• Pick your best result as the reference. Use your strongest image as the character reference for the next ones.
• Expect to retry. Generating several versions and choosing the closest is a normal part of the process.

• Very long prompts. Too many details compete and some get ignored. Keep the words that matter.
• Mixed styles. "Anime, photorealistic, oil painting" in one prompt gives muddled results.
• No lighting. Lighting changes a portrait more than almost anything else, so always name it.
• Using someone's photo without permission. Only use reference photos of yourself or people who agreed.

The NanoPrompts generators build the subject, lens, lighting and setting parts of a prompt for you. They include wording that asks the AI to keep the same person, which helps but does not guarantee identical faces. Copy the prompt into Midjourney, then add your own settings such as --ar and --cref.