With multimodal abilities, I wonder if it could eventually check a raster of the SVG it created and then make iterative improvements that get a better result th

With multimodal abilities, I wonder if it could eventually check a raster of the SVG it created and then make iterative improvements that get a better result than vectorizing a raster from stable diffusion.