← Back to Blog

10 Tips for Writing Better Midjourney Prompts

I've generated a genuinely embarrassing number of Midjourney images at this point, most of them mediocre, some of them exactly what I wanted, and a few of them better than what I could've described even if I'd tried harder. Somewhere along the way, a handful of small habits made the difference between images that felt like happy accidents and images I could actually reproduce on purpose. None of this is complicated, but it took longer to figure out than I'd like to admit. Here's what actually moved the needle, not the generic advice you see copy-pasted everywhere.

1. Lead with the subject, not the mood

It's tempting to open with atmosphere, something like "a moody, cinematic..." before you've even said what the image is of. The problem is that Midjourney tends to weigh earlier words more heavily than later ones, so leading with vibe instead of substance means the actual subject gets less attention than it deserves. Start with what the image is actually of, get that locked in first, and then layer the mood, lighting, and style on afterward. It sounds like a small reordering, but it noticeably changes how consistently the subject shows up the way you pictured it.

2. Vague style words waste your prompt

"Digital art" tells the model almost nothing useful, there are a thousand different things that phrase could mean, and the model has to guess which one you're picturing. Naming something specific instead, like Ukiyo-e woodblock print or 1980s airbrush illustration, gives it an actual visual reference to work from instead of a vague category. The more specific and namable the style, the more consistent your results tend to be across different subjects. This was probably the single biggest quality jump for me early on.

3. Lighting does more than people expect

I underestimated this for way too long, treating lighting as an afterthought rather than a real lever. Two otherwise identical prompts with just the lighting phrase swapped, golden hour lighting versus dramatic chiaroscuro, will come back looking like they belong to completely different projects. It affects color grading, mood, and even how detailed the image feels, far more than most people expect from a couple of extra words. If you only add one new habit from this whole list, make it deliberately choosing a lighting phrase every time instead of leaving it to chance.

4. Say where the camera is

Phrases like low-angle shot, wide establishing shot, or close-up framing push the model toward an actual, specific composition instead of defaulting to whatever centered, medium-distance pose it usually falls back on. Without this, a lot of results start to look the same after a while, centered subject, neutral distance, safe framing. Naming a camera position is a fast way to break that pattern and get compositions that feel more intentional and varied.

5. Commas beat sentences

This one felt counterintuitive at first, since we're trained to write in proper sentences everywhere else. But Midjourney parses a list of descriptive phrases separated by commas more effectively than a grammatically complete sentence, it's reading for concepts and keywords, not for correct grammar. Writing "a cat, sitting on a windowsill, soft morning light, shallow depth of field" tends to outperform the same idea written as a full sentence with connecting words.

Write it like a list of ingredients, not a story.

6. Add something for quality, but don't overdo it

Phrases like highly detailed or sharp focus nudge the output toward more refined, polished results, which is genuinely useful. Where people go wrong is stacking four or five of these quality phrases in a row, assuming more must mean better. In practice it doesn't multiply the effect, it just eats up space in your prompt that could've gone toward describing the actual subject, style, or composition more precisely. One or two quality phrases is usually plenty.

7. Mood words matter even in non-people scenes

Words like eerie, serene, or triumphant shift color grading and atmosphere noticeably, even in a landscape or object scene with no person and nothing overtly emotional happening in it. I didn't expect this to matter outside of portraits, but it turns out mood words affect the overall rendering style broadly, not just facial expressions or body language. Worth testing on a scene you'd assume doesn't need it, the difference is usually bigger than expected.

8. Don't fight yourself with contradictions

"Minimalist" and "highly detailed" in the same prompt send the model mixed signals, and the result usually ends up an uneasy compromise between the two instead of confidently being either one. Pick one direction per prompt and commit to it. If you're genuinely curious how the other approach would've looked, generate a second version with that direction instead of trying to blend both into a single confused result.

9. Treat the first result as a draft, not a verdict

A weak first output usually isn't the model failing, it's information about what to adjust next. Instead of scrapping the whole prompt and starting over, look at what specifically went wrong, the composition, the style, the lighting, and change just that one or two things. This iterative approach gets you to a result you're happy with noticeably faster than repeatedly writing brand new prompts from scratch each time.

10. Keep the combinations that work

When a particular style, lighting, and composition combination clicks and produces something you genuinely like, write it down somewhere you'll actually find it again. It usually transfers surprisingly well to a completely different subject too, which saves a lot of time you'd otherwise spend reinventing a good combination from scratch every time you start a new image.

If you'd rather not build all this from scratch every time, the PromptNest generator handles the composition, lighting, and detail phrasing for you automatically in Image Prompt mode, you just describe the subject and pick a style.