Figuring out the midjourney prompt basics
The first time I tried to get Midjourney to make flat design icons, I didn’t even think about keywords. I just typed something like “flat icon of a computer” and hit enter. What I got back was sometimes fine but more often it was textured, shaded, or had gradients I didn’t want. It looked more like clip art from a cheap presentation template than the clean vector feel you’d expect from flat design. The trick, I found out, is to start stacking descriptive keywords instead of hoping the AI understands minimalism by default.
When I started adding words like “flat,” “vector,” “solid fill,” and even “no gradient,” the generations shifted. Suddenly the little icons had thick outlines or monochrome fills, much closer to what most icon packs look like. If you’re completely new, think of keywords as reminders you are whispering to Midjourney about what NOT to do. If you forget “no shadow,” you’ll get a little drop shadow on half your results and spend another half hour trying to re-roll.
How I learned to force consistency in sets
Trying to create one flat icon is easy enough, but when I wanted a whole set — like a folder, a magnifying glass, and a phone — every single one came out in slightly different proportions. One icon had a thick line weight, another had barely visible outlines. Next thing I know I’m dragging everything into Figma and manually resizing, which is not the dream. The way I fixed it was to bake the style into the keywords. I got more consistent results using “2d flat icon vector minimal consistent line weight simple symbols.”
Also, when you are making multiple prompts, keep the prompt wording almost identical across generations. Change only the subject, not the style words. For example:
– “2d flat vector icon solid fill folder”
– “2d flat vector icon solid fill phone”
– “2d flat vector icon solid fill magnifying glass”
That way Midjourney treats the subjects differently but locks the same design language across all three.
Debugging when shadows sneak back in
Even when I swear I had the right keywords, shadows often crept in out of nowhere. I would put “no shadows” at the end, but the icons still showed faint gray borders suggesting dimension. It drove me crazy. What finally helped was moving “no shadows no gradients” up to the start of the prompt. For some reason Midjourney sometimes ignores the negations if they are buried at the bottom. When I put them first, the AI respected it more often.
Another hack was adding “simple flat pictogram style” — pictogram makes it think of the kind of icons you see on public signs, which never have shadows or fancy lighting. If you find one batch still sneaks them in, don’t waste endless re-rolls. Just try another variation of the same negation keyword. I’ve alternated between “no shadows” and “without shadows” and seen different results.
Tricks for achieving transparent backgrounds
Midjourney always insists on putting a background behind icons, whether it is white, gray, or some weird gradient. And since I actually wanted vector icons I could drop on any color, this became another messy detour. The phrase “icon isolated on white background” actually worked better than “transparent background.” Once I had it on clean white, I just removed the background later in Figma or Photoshop. If you try “transparent background” Midjourney will generate some weird checkerboard texture because it tries to literally draw transparency rather than outputting PNG alpha.
The best workflow I landed on was to request “flat icon vector pictogram isolated on plain white background no shadows no gradients.” Then I batch removed the backgrounds. It is clunky but reliable. If you are making multiple icons for an app, you will save your sanity by enforcing one solid background in the prompt rather than hoping for literal transparency.
When colors get too playful
Flat doesn’t mean garish, but Midjourney sometimes thinks otherwise. Give it no restriction and suddenly you have neon pink printers and bright green phones. If you want muted consistent colors, you need to tell it. Words like “monochrome,” “duotone,” or “limited palette” help a lot. I often put “flat vector icon minimal duotone color scheme white and dark blue” and then just swap different color keywords for new sets.
It’s a little like tutoring a toddler. Saying “flat design” alone doesn’t always prevent chaos. You need to tell it how many colors should exist, what kind of vibe those colors have (warm, muted, pastel), and sometimes the exact named colors. Midjourney oddly understands “Pantone” too, although results vary.
Scaling issues and output sizes
Probably my least favorite part is resizing. Icons spit out of Midjourney come with huge square compositions. If you crop them too closely, the details get messy. Asking for “vector style scalable flat icon” gives cleaner shapes that survive scaling better. It does not make a true vector file (you still get a raster), but when you open it in Figma the edges trace smoothly if you run auto-trace or pen tool adjustments.
For practical use, I started requesting them bigger and then downscaling instead of trying to enlarge them. Midjourney actually outputs higher detail with “–uplight” in the command, and that’s helped when I need tiny 16px versions that don’t look like blobs.
Keyword combos that broke things anyway
Sometimes adding too many clarifications has the opposite effect. Once I tried “2d flat minimal pictogram icon no shadows no gradients vector white background limited color duotone simple illustrative scalable universal easy to understand symbolic.” The poor thing spit out abstract art blobs that looked like someone spilled ink. Too much input confuses it. Keeping the wording shorter and repeating only the style phrases you care about gets better outcomes.
The best balance for me has been three style cues plus one or two content words. Something like: “flat vector icon no gradients folder” is usually enough. If you try to micro manage, you’ll spend hours rerolling frustrated that every one looks like junk.
Where to keep experimenting with styling
If you want reference packs to compare, I grabbed some random free icon sets online to see how their flat styles were phrased. Sites like flaticon.com are useful for browsing. Even if you don’t use them, just scrolling sets helps you figure out what “line based” vs “filled” actually looks like in practice, so you can pick keywords accordingly. Otherwise you’re stuck guessing, and Midjourney thrives on solid descriptive words instead of vague expectations.
Once you start thinking like a designer giving instructions to a very literal intern, the keywords almost write themselves. I learned to treat every failed render as a clue for a new word to add or remove. Honestly, you’ll spend more time formatting prompts than actually downloading icons, but that’s half the fun 🙂