App icon prompts. Readable at 32 by 32.
Pick a category, glyph type, style era, color treatment, and background; get a Midjourney, Nano Banana, or Imagen-ready prompt with explicit aspect, simplicity directive, and iteration count. Browser-only; nothing is saved.
App, glyph, era, color, background.
Your icon prompt.
Click build to generate.
What good looks like.
Notewell: a notes app where the icon is a single curved leaf abstracted into the letter N
Glyph type: abstract. Era: Apple flat. Color: flat solid in moss green (#3F5947). Background: subtle warm cream gradient. The leaf reads at 32x32 because the silhouette is a single closed shape; the negative space inside the leaf forms the negative of the letter N. Generate 4 variations in Midjourney v6.1 at 1:1 aspect, then export the strongest at 1024 master and run an SVG vector trace for the App Store binary.
Pawday: pet-tracking app, mascot is a stylized paw print rendered in soft claymorphism
Glyph type: mascot. Era: claymorphism. Color: gradient duotone (warm coral to dusty rose). Background: flat off-white. The 3D clay rendering gives the paw print a tactile, fingertip-friendly feel - critical for a pet-care app where the audience leans warm-emotional rather than utility-clinical. Best generated in Anthropic tooling or DALL-E 3 because both handle clay rendering more reliably than Midjourney.
Tally: investing and stock-tracking app, glyph is an upward chevron in gradient duotone
Glyph type: geometric primitive. Era: Apple flat. Color: gradient duotone (deep navy to electric green - the universal "up" colors of finance). Background: matching navy gradient. The chevron is 60% of the canvas; the icon reads as "rising" at any size. This is the canonical finance-app icon pattern and is intentionally generic - in finance, recognizability over distinctiveness wins because users sort by trust signal first. The icon should pair with a strong wordmark when shown alongside competitor apps in Figma mockups.
An icon is a silhouette, not a scene.
An app icon ships at sizes from 1024 px down to 29 px. The 29 px size is what determines whether anyone finds your app on a home screen. Most AI-generated icon attempts fail because the prompt asked for a "scene" (a person using the app, a clever visual metaphor with three elements, a tiny illustration) rather than a silhouette. A good prompt names the glyph as a single shape, fixes the era, names the color, and adds the simplicity directive: "must be readable at 32 by 32 pixels." Without that line, the model will pack detail it should have left out.
Readable at 32x32 is the contract
The single most important constraint on any app icon is that it must remain identifiable at the smallest size the OS will render it at. On iOS that is the Settings list and Spotlight (29 px and 40 px respectively). On Android it is the launcher app drawer at varying sizes per OEM. The simplicity directive in the prompt - "the glyph must be readable at 32 by 32" - forces the model to drop detail that would muddy the silhouette. After generation, do the squint test: blur the icon to 8 px and check whether the silhouette is still distinct from competitor apps in the same category. Apple's Human Interface Guidelines are the canonical reference here.
Glyph not scene
The single most common AI-icon failure is "scene" composition - a tiny person, a tiny laptop, three small icons inside the icon. None of those scale. The fix is to constrain the prompt to a single glyph: one shape, one concept, one focal point. If the app does five things, the icon should not depict all five; it should depict the one that matters most or the abstract symbol that contains all of them. Compare the App Store top-100 by category: nearly every winning icon is a single-shape glyph (Spotify circle with three lines, Notion N, Slack four-color hash, Figma four colored shapes). The pattern is consistent.
One concept per icon
Related to glyph not scene: the icon must say one thing. A pet-care app icon does not also need to communicate "trustworthy" or "fast" or "loved by Gen Z" - those go in the App Store screenshots and metadata, not the icon. The icon's job is to be findable on a home screen full of competing icons. If the user can identify your app on first glance, the icon has done its full job. Everything else is decoration. Reference: web.dev articles on visual hierarchy.
Color reads at thumbnail scale
The default AI-generated palette is too desaturated and too muddy. Real-world App Store icons use saturated, contrast-heavy color because that's what reads at thumbnail. Test by saving the generated icon as a 60 px PNG and previewing it next to your competitor apps in a screen mockup; if your icon visually recedes, push the saturation. Duotone gradients work well at thumbnail because the eye sees direction (top-to-bottom hue shift) before it sees detail. Brand-led palettes work when the brand color is already saturated; they fail when the brand uses a muted palette designed for print rather than screen. Use the color contrast checker to verify the icon shape passes 3:1 contrast against both light and dark home-screen wallpapers.
Test the dock contrast
The App Store icon is shown on a white-ish background. The home-screen icon sits on the user's wallpaper - which is rarely white. Test the icon on three backgrounds: pure white (App Store), pure black (Android dark mode), and a busy photo wallpaper. If the icon disappears against any of those, it needs a more contrasted background or a defined edge stroke. The "negative-space bleed" background option is risky because it relies on the icon shape extending to the edge - which works on iOS rounded squares but can collide with adaptive-icon masks on Android. When in doubt, use a flat or gradient background that defines a clear border.
Iterate at the cheapest fidelity
The biggest mistake is generating one icon at high fidelity (a fully rendered 1024 px PNG) and then trying to redesign it. Better: generate 9 to 16 variations at the start, eliminate the bottom 80% in 30 seconds (squint test, dock test, competitor-side-by-side), then take the top 2 to 3 for high-fidelity refinement. The cheapest fidelity is the prompt itself - if you do not like the first batch, change the glyph type or the era and regenerate; do not try to refine a bad glyph into a good one. Patterns that come from Stability AI tooling generally need fewer iterations than purely free-form prompts.
Related work: Brand identity for full app brand systems. Favicon generator for the web counterpart. Website hero prompt generator for the matching landing page hero. Business name generator if the app still needs naming. Web design for the marketing site that introduces the icon to first-time visitors.
Five answers.
Should I use Midjourney, Nano Banana, or DALL-E for app icons?
Each handles a different style well. Midjourney v6.1 is strongest for flat, geometric, and brutalist eras; outputs feel like a graphic designer made them. Nano Banana (Gemini's image model) is currently the strongest for soft claymorphism and 3D-rendered icons because it preserves form better. DALL-E 3 is the most flexible for hand-drawn and vintage styles. None of the three reliably produces App Store ready output without a vector cleanup pass - always SVG-trace the winning raster output before submitting to App Store Connect or Google Play.
Why does the prompt lock to 1024x1024?
App stores require a square master at 1024x1024 (App Store Connect) or 512x512 (Google Play). Generating at the master size lets you scale down without artifacts. Generating non-square output and then forcing it square produces a clipped or distorted glyph. The 1024 master also gives you headroom for the dozen smaller sizes both stores need (Spotlight, Settings, Notification, App List). If you only have a 512 px output, your Notification icon at 60 px will look soft.
What's the difference between glyph types?
Abstract symbols (Spotify's three lines, Slack's hash) work for products that span many features - the abstract shape becomes the brand. Single-letter monograms (Notion's N, Discord's D) work when the brand name is already memorable. Silhouette objects (Calculator's calculator shape) work for utilities where the function is the brand. Geometric primitives (chevrons, hexagons) work for finance and trading. Hand-drawn lines work for journaling, creative, and craft apps. Negative space works rarely - it requires a strong shape and is hard to read at thumbnail. Mascots work for kids' apps, pet apps, and games but are the highest-skill option to execute well.
Do I need a designer if the AI gives me good output?
For early validation and prototype builds, the AI output is enough. For App Store submission, you need a vector cleanup pass: re-trace the raster as an SVG so the icon is resolution-independent, fix the corner radius to match Apple/Google specs, generate the full size set (29, 40, 60, 80, 87, 120, 180 for iOS plus mipmap-mdpi through mipmap-xxxhdpi for Android). The AI gives you the design; a designer or design-aware engineer ships the asset. This pattern is the cheapest path to a defensible icon - design via AI, productionize via vector tools.
Does this tool save my icon prompts?
No. Every value you enter and every prompt that gets built lives in your browser tab only. Nothing is transmitted to a server, stored in a database, or synced across devices. Close the tab and the data is gone. The Copy prompt button puts the assembled prompt on your clipboard; that is the only output path.
Icons are silhouettes.
Our app-brand engagements pair an icon system with the wordmark, app store screenshots, and adaptive Android variants. Two-week turnaround on the icon, vector master, and full size set.