Generating a 1024x1024 image with StableDiffusion 3 in KerasHub, using float16 inference.
The `generate` method is compiled, and compilation happens on the first call. The next time you call the model again you will reuse the same compiled function and it will be super fast.