Midjourney is one of the most powerful AI image generators available today. The images it produces can be stunning, surreal, photorealistic, or painterly — depending entirely on how you write your prompts.
The problem most beginners run into: they type something simple like “a dog on a hill” and get a mediocre result, then assume the tool isn’t that impressive. The truth is, Midjourney rewards prompting skill. Learn a few core techniques and your results go from average to genuinely remarkable.
This guide covers everything you need to get started.
Table of contents
Open Table of contents
How Midjourney Works (The Very Short Version)
Midjourney runs inside Discord. You join their server (or use the Midjourney website), type a command in a channel, and the AI generates four image variations based on your description. You can then upscale your favorite, generate variations of it, or start over with a refined prompt.
To access Midjourney, go to midjourney.com and sign in with Discord. There’s no fully free tier anymore — the Basic plan starts at around $10/month and gives you about 200 image generations per month.
Once you’re in, use the /imagine command followed by your prompt.
Anatomy of a Good Prompt
A Midjourney prompt isn’t just a description — it’s a set of instructions that guides the AI’s interpretation of style, mood, subject, and composition. Here’s a simple structure that works well:
[subject] + [setting/context] + [style/medium] + [mood/lighting] + [parameters]For example:
a lone lighthouse on a rocky coastline, stormy sky, oil painting style, dramatic side lighting, moody --ar 16:9 --v 6.1
Compare that to just typing: lighthouse. Same subject, very different results.
The Most Useful Parameters
Parameters are added to the end of your prompt with two dashes. You don’t need to memorize all of them — just these five will cover most situations.
--ar (aspect ratio)
Controls the shape of your image. Common values:
--ar 1:1— square (good for social media)--ar 16:9— widescreen (good for banners and backgrounds)--ar 9:16— portrait (good for phone wallpapers)--ar 4:3— traditional screen format
--v (version)
Specifies which Midjourney model to use. As of 2025, --v 6.1 is the current default and produces the best results for most prompts.
--style raw
Reduces Midjourney’s tendency to add its own “artistic flair.” Use this when you want a cleaner, more literal interpretation of your prompt — useful for realistic photography or product images.
--chaos (0–100)
Controls how varied the four generated images are. Low chaos = four similar images. High chaos = four very different interpretations. Try --chaos 30 when you want to explore options.
--no
Tells Midjourney what to exclude. Example: --no text, watermark, people removes those elements from the image.
Prompting Techniques That Actually Work
Be specific about style and medium
Instead of “a painting of mountains,” try:
snow-capped mountain peaks at sunrise, impressionist oil painting, thick visible brushstrokes, warm amber and violet tones
Style references work especially well. Try adding: in the style of [art movement] — like “in the style of Art Nouveau” or “in the style of 1970s vintage travel poster.”
Describe lighting explicitly
Lighting is one of the biggest factors in how professional an image looks. Some useful terms:
golden hour lighting— warm, low-angle sunlightsoft diffused light— even, shadow-free lighting (good for portraits)neon-lit— colorful artificial lightchiaroscuro— strong contrast between light and darkbacklit— subject lit from behind, creates silhouette or glowing edges
Reference camera or lens settings for photorealistic images
If you want something that looks like a real photograph:
portrait of a woman in a café, shallow depth of field, shot on 85mm lens, natural window light, film grain
Terms like “shot on Sony A7”, “35mm film”, “DSLR”, and “macro photography” all push Midjourney toward more photorealistic outputs.
Use weighted emphasis
You can put double colons :: after any word to increase or decrease its importance:
ocean::2 sunset::1 sailboat::0.5
This tells Midjourney to emphasize the ocean most, the sunset somewhat, and treat the sailboat as a minor element. Useful when one part of your scene keeps getting ignored.
Example Prompts to Try
Here are some ready-to-use prompts across different styles:
Realistic portrait:
close-up portrait of an elderly fisherman, weathered face, deep-set eyes, overcast coastal light, Leica photograph, sharp focus, documentary style --ar 4:5 --style raw
Fantasy scene:
ancient library inside a giant crystal cave, glowing bioluminescent mushrooms, floating books, ethereal atmosphere, digital art --ar 16:9 --v 6.1
Product photography:
minimalist perfume bottle on white marble surface, soft studio lighting, clean background, luxury editorial style, commercial photography --ar 1:1 --style raw
Social media graphic:
modern flat design illustration, productivity and focus concept, pastel colors, geometric shapes, clean lines, app icon style --ar 1:1
Travel poster:
vintage travel poster for Kyoto Japan, cherry blossoms, Mount Fuji, bold typography space, retro 1960s illustration style, limited color palette --ar 2:3
Common Beginner Mistakes
Being too vague. “A nice landscape” gives Midjourney almost nothing to work with. The more specific and layered your description, the better.
Ignoring aspect ratio. The default is a square. If you’re making a banner, wallpaper, or social post, set the right ratio from the start or you’ll crop awkwardly later.
Not iterating. Midjourney rarely produces your ideal image on the first try. Pick the closest result from the four options, click “Vary (Subtle)” or “Vary (Strong)”, and keep refining. Most great images take 5–10 generations.
Adding too many conflicting ideas. A prompt with five different styles, four different moods, and three subjects will confuse the model. Focus on one clear concept per prompt.
How to Get Consistent Characters or Styles
One of the more advanced techniques is keeping a consistent character or style across multiple images — useful for illustrations, branding, or storytelling.
The easiest way: use --sref (style reference) followed by a URL of an image you’ve already generated. This tells Midjourney to match the visual style of that reference image.
For character consistency, you can use --cref (character reference) the same way — point it at a face you’ve already generated and Midjourney will try to keep that character consistent in new generations.
Getting Better Over Time
The best way to improve is to study what works. Midjourney’s website has a public gallery of community images — when you see something you love, click it to see the prompt. This is the fastest way to learn which words and combinations produce strong results.
Keep a personal prompt library. Copy prompts that work and tweak them for new subjects. Over time you’ll build a personal “vocabulary” that consistently produces images you’re happy with.