Ask HN: Could the Apple Neural Engine be used to run Stable Diffusion?

I don't really know how to look at neural engine specs and figure out if it can run CUDA (probably can't right?) but if the bindings existed and the Neural Engine interface was public (apparently its private and can only be used by Apple or via the MLkit offering) could it actually do the matrix mathemagics needed to run say txt2img or img2img natively?