AFAIK that model is pretty old, and it was explicitly trained for SVG generation. For other models the capability of generating SVGs of real stuff is accidental

AFAIK that model is pretty old, and it was explicitly trained for SVG generation. For other models the capability of generating SVGs of real stuff is accidental. Same as GPT-5.x and Sonnet 4.5+ being able to generate MIDI music.