An AI model learns icon styles from examples, not from written instructions. Providing two high-quality examples for each concept helps the AI model generate better icons. A two-step process, first generating concepts then generating icons, is efficient.
By Pawan Nayar · from the LLOS AI Learning record
An AI model learns icon styles from examples, not from written instructions.
Providing two high-quality examples for each concept helps the AI model generate better icons. A two-step process, first generating concepts then generating icons, is efficient.
The AI model needs good examples to learn from. Do not give the AI model general instructions like 'use basic SVG elements'. These instructions do not help the AI model learn style. The AI model learns the desired style from the example icons themselves. This method produces much better results than written descriptions. Use a two-step process. First, the AI model creates many ideas for icons. It also writes a short description for each idea. Second, the AI model uses these ideas and two example icons to create the final icons. This two-step method is called a '2-hop pipeline'. In the first step, the AI model takes a general topic. It then creates 100 different icon ideas. Each idea includes a short visual description. In the second step, the AI model makes 100 separate requests. Each request uses two example icons and one idea from the first step. For each idea, the AI model creates three different icons. This creates 300 icons in total. Many requests can run at the same time.
A developer wants to create a set of weather icons. The developer gives the AI model two examples of existing weather icons in the desired style. The developer does not tell the AI model to 'use simple shapes'. The AI model then generates 100 weather-related concepts. For each concept, the AI model creates three new icons based on the two examples. This results in a large set of icons that all share a similar style.
Resolved from this lesson's own record — real definitions, not invented for this page.
More from the same record, pulled by shared topic — not written new for this page.
The system uses 50 hand-crafted noun silhouettes. These are organized into five domains.
All parts of an interactive chart must connect smoothly. This means shapes should share edges or points.
A developer must define strict rules for SVG attributes. This includes setting `viewBox`, `width`, `height`, and `role` for accessibility.
The AI model acts as a technical SVG generator. It uses two colors: 'fg' for the main body and 'ac' for one important feature.
An AI model learns icon styles from examples, not from written instructions. Providing two high-quality examples for each concept helps the AI model generate better icons. A two-step process, first generating concepts then generating icons, is efficient.