That wasn't the prompt. That text was generated by asking the model to describe an image and feeding it a rendering of the SVG it had previously generated.

No, the prompt I always use is "Generate an SVG of a pelican riding a bicycle".