Using descriptive text instructions to guide or control how a model generates output, such as specifying desired voice characteristics.