Generating a Matching Icon Set in Leonardo AI Without a Designer
Eleven icons, one visual language. The trick is not the prompts, it is building the first icon until it is perfect and then never changing that prompt.
My side project needed eleven feature icons and my budget needed them to cost nothing. Leonardo AI produced the set in an afternoon, and the workflow that made it work was one specific discipline: get the first icon perfect, then treat its prompt as scripture.
Phase one is entirely spent on icon one. Style anchor (I used 'flat vector icon, two-tone, single accent color'), explicit line weight, explicit corner rounding, plain background, no text. I generated maybe twenty variants, picked one, and noted every style phrase in the prompt. This first icon is the style guide; everything else is derived from it.
Phase two is the cloning phase, and the rule is brutal: change only the subject noun. 'Briefcase' becomes 'chart' becomes 'shield.' The moment you 'improve' a later icon's prompt, you break the set, because consistency comes from identical instructions, not from similar-looking intentions. Where a subject fights the style, I regenerate the subject prompt, not the style block.
Two Leonardo features carried real weight. The image guidance feature lets you feed icon one back in as a style reference for the later generations, which rescued the three subjects whose natural look fought the style. And the transparent PNG export meant no background removal pass at all.
Honest scorecard: two of eleven icons needed a manual pass in a free vector editor to thin out a stroke, and none are as conceptually clever as a designer's work. But they are consistent, on-brand, mine, and cost an afternoon. For an MVP, that trade is not close.
