but noise textures let you "suggest" as shape as well
you can think of a fuzzy noise texture as a probability distribution for where a shape it likely to appear

you see the gradient? they're neither 1 (white) or 0 (black). this is one of the way shaders can generate very fluid shapes, by overlaying multiple noise textures, and the principle can be applied elsewhere too (inc procedurally generated noise, such as from a Perlin noise function)
here's a talk to see what I'm talking about
timestamp @ 25m, he scrolls two noise textures over one another, and the result of scroll two "probability distributions" over one another is different possibilities pop into existence, producing smooth clouds
the same principles can be applied to other noise-based generation