Turning a Designer's Imagination into Reality: Building Yeogieottae's Icon Generator

As the service expanded, demand for icons exploded, but the process of creating them while maintaining YDS (Yeogi Design System)'s strict brand guidelines (translucent plastic texture, bevel treatment, 3D angles, brand colors) was a serious bottleneck. Generic AI tools produced inconsistent styles every time, making them unusable for practical work.

They solved this problem with a multi-view split generation strategy. By designing prompts to split the canvas left and right—placing a 3D view on the left and a shadowless 2D flat style on the right—they obtain an icon set in both styles from just a single generation. When users attach a sketch or reference image, the shape is preserved while only the YDS style is applied.

The real core is the pipeline that converts bitmaps into editable vectors.

• Force the background to pure white (#FFFFFF) and prohibit pure white inside the subject → remove boundary noise using a pixel scanning algorithm
• Instead of placing hundreds of anchor points like external vectorization tools, compress into clean plane data that looks hand-drawn through path merging, coordinate cleanup, and outline removal
• Results are copied to the clipboard instantly—PNG for 3D, SVG for 2D—so they can be pasted directly into Figma

With an infinite canvas UI, users can freely explore for inspiration without pagination, save favorites, and batch download them as a ZIP.

Key takeaway: AI requires continuous maintenance. Prompts must be retuned with every model version upgrade, and only through repeated fine-grained UX improvements that integrate into practical workflows does it become a 'living service.'