Midjourney Prompts: What I Actually Learned After 500+ Images

I've been messing around with Midjourney for about six months now. Somewhere around image 500, something clicked — I stopped thinking about prompts as keyword lists and started thinking about them as scenes I'm describing to a painter who can't read my mind. That shift changed everything.

This isn't a parameter reference. You can find those anywhere. This is just me sharing what I actually learned from making a LOT of mediocre images before the good ones started showing up.

The One Thing That Changed My Prompts

A prompt is not a list of keywords. It's a scene.

When I typed "a cat," I got a cat. Average, forgettable, basically the same thing you'd get from an image search. But when I typed "a fat orange cat with blue eyes wearing a red scarf, sitting on a winter windowsill while snow falls outside, warm indoor lighting, Miyazaki-style healing illustration" — I got something I'd actually want to keep.

The difference isn't clever keyword usage. It's specificity of imagination.

A structure that works for me: subject + details + environment + style + lighting + composition + parameters. You don't need every element every time. But when I'm stuck, I walk through each category in my mind and fill in what matters. The prompt kind of writes itself.

Parameters I Actually Use

There are dozens of Midjourney parameters. I use maybe five on a regular basis.

--ar (aspect ratio): This is the most underrated parameter, and almost every beginner ignores it. They leave the default 1:1 square every single time. But aspect ratio dramatically changes how an image feels. Portraits look better at 2:3. Landscapes at 16:9. Cinematic scenes at 21:9. A cyberpunk city at 1:1 versus 21:9 literally feels like a completely different image.

My go-to ratios:

• People and phone wallpapers: --ar 2:3 or --ar 9:16
• Landscapes and cinematic shots: --ar 16:9 or --ar 21:9
• Social media headers: --ar 3:1
• Instagram vertical: --ar 4:5

--s (stylization): This controls how much creative freedom Midjourney has. Low values (around 100) follow your description closely. High values (750 and up) go wild artistically.

My habit: I use --s 500 for exploration — lets Midjourney surprise me. Then I drop to --s 100 when I'm fine-tuning a specific image. Going all the way to --s 1000 sometimes produces stunning results, but sometimes it's so out there I can't even use it.

--style raw: This one matters if you want photorealistic images instead of "artistic" ones. It makes Midjourney behave more like a camera and less like a painter. Combine it with --s 100 for best results.

Something like:

portrait of a 25-year-old woman, natural light, by a window, Sony A7R4 85mm f1.4, skin detail --ar 2:3 --v 6 --style raw --s 100

--no (negative prompts): Things you don't want. I almost always include "--no deformed, ugly, extra fingers, watermark, text." It's not perfect — you'll still get the occasional weird hand — but it cuts down on the most common issues a lot.

--seed: When you get an image that's close but not quite right, note the seed value. Reusing the same seed with modified prompts lets you iterate on composition without starting from scratch. I keep a small note file where I log seed numbers for images I liked. Sounds obsessive, but it saved me hours of "I had something close before but I can't recreate it."

Three Things That Actually Leveled Up My Prompts

1. Replace abstract words with concrete ones

"Beautiful," "gorgeous," "high quality" — these mean essentially nothing to Midjourney. They're filler.

Instead of "a beautiful photo," write "shallow depth of field, soft bokeh, golden hour light, Leica M10 color tone." Instead of "a cool cyberpunk scene," write "rain, neon reflections on wet pavement, holographic ads, steam, dark moody lighting."

Specific descriptions beat adjectives every time. Every single time.

2. Artist names are style shortcuts

Why spend fifty words describing a style when one name does the job?

"Miyazaki Hayao style" — Midjourney knows exactly what you mean. "Makoto Shinkai style" — that specific luminous sky feeling. "Ansel Adams style" — black and white large-format landscape texture instantly.

These aren't tricks. These artists have such distinctive visual signatures that their names are essentially style codes.

Some I use frequently:

• Photography: Ansel Adams for landscapes, Annie Leibovitz for portraits, Henri Cartier-Bresson for street
• Illustration: Miyazaki, Makoto Shinkai, Alphonse Mucha
• Painting: Monet, Van Gogh, Vermeer

I also lean on color palette names sometimes — "Morandi color palette" gives this muted, sophisticated look that I love for certain moods.

3. Image-to-image is my real workflow

Pure text descriptions sometimes can't precisely convey the picture in my head. So my actual process is:

1. Quickly generate a batch with --q 0.5 to save time
2. Find one with potential composition
3. Use it as a reference image with --iw 1.2
4. Adjust the text description and iterate

--iw controls reference image weight. Between 0.5 and 1.5 works best. Too high and Midjourney gets locked into the reference. Too low and it won't matter.

Good Images Take Iteration. A Lot of It.

My first image is never my final image. Never.

My standard process looks like this:

1. Explore: Generate a batch quickly (--q 0.5), scan for promising directions
2. Select: Find one with potential, note the seed
3. Refine: Adjust the prompt, tweak parameters, use Vary to explore variations
4. Finalize: When it's close, switch to --q 2 for maximum quality

A genuinely good image takes me 10-20 rounds of iteration on average. That's not an exaggeration. The "one prompt, perfect image" thing you see on social media is mostly luck, or they're not showing you the 99 failed attempts.

This was the hardest lesson for me to accept. I kept thinking I was doing something wrong because my first tries weren't good. But that's just how it works. Iteration is the process, not a sign of failure.

Vary Region: The Feature That Changed Everything

Midjourney V6's regional variation lets you select part of an image and regenerate just that section.

This solved probably 80% of my "almost perfect but..." problems. Great portrait but the hands came out wrong? Select just the hands and regenerate. Composition is good but the background is weird? Fix the background without touching the subject.

Before this feature, a single bad element meant regenerating the whole image and hoping everything else stayed good. It was genuinely frustrating. This made it manageable.

My Go-To Prompt Fragments

Over time, I've collected phrases that produce reliable results. I mix and match them depending on what I'm going for:

Lighting:
golden hour light, volumetric light, rim lighting, Rembrandt lighting, neon glow, candlelight, softbox

Camera/Lens:
Sony A7R4 85mm f1.4, Leica M10 35mm f2, Hasselblad X2D 90mm, shallow depth of field, bokeh, film grain

Mood:
serene, melancholic, epic, mysterious, nostalgic, hopeful, unsettling, healing

Materials:
brushed metal, weathered leather, frosted glass, silk, aged wood, rust, translucent

Styles:
Miyazaki style, Makoto Shinkai style, Blade Runner style, Ansel Adams style, Morandi color palette, cyberpunk tone

These aren't magic words. They're just vocabulary that's produced consistent results for me. You'll build your own library as you go.

Common Mistakes I Still See (and Used to Make)

Contradictory prompts. "Realistic cartoon style" — pick one. The model will try to do both and fail at both. I've done this more times than I'd like to admit.

Too many concepts. Every element you add dilutes the others. A prompt about "a cyberpunk samurai cat princess in a neon forest with dragons and a castle" is going to be a mess. Focus on what matters most. You can always add elements in later iterations.

Giving up too soon. Midjourney has a randomness factor. Sometimes you need to reroll a few times before the magic happens. I've had prompts that produced garbage the first three times, then generated something stunning on the fourth try. Don't judge a prompt by one result.

Ignoring lighting. This was my biggest blind spot early on. Lighting is what makes an image feel professional versus flat. If something looks "off," try adding specific lighting descriptions before changing anything else. It fixes more problems than you'd expect.

One Last Thing

Parameters and techniques are just tools. The thing that actually determines how good your images are is your eye — what you've seen, what you find beautiful, what you're trying to express.

Look at good photography. Watch movies and pay attention to how scenes are lit and composed. Study paintings. Visit art galleries, scroll through photography portfolios, pay attention on the street.

The more visual input you take in, the better your prompts become — because you'll have more specific things to ask for. Midjourney is the tool. Your taste is what makes the art.

Technique can learn. Taste has to be cultivated.