No it's not. Generating an SVG is asking the model to write text that can be rendered as an image.
Generating an image directly skips the intermediate step and directly outputs an image: so the accuracy is in a completely different league. The models people ask for SVGs typically cannot do this, only specially post-trained variants can.
(An LLM will do the svg of a pelican on a bike much more accurately btw.)
The medium is superfluous.