Your value proposition is that the user doesn't have to think about positioning and other stuff. This is a great optimization for certain use-cases, but it will definitely diminish your addressable market. Many diagramming use-cases need the additional customization that positioning provides.
One might think there's an algorithm that can produce perfectly optimal layout for any declaration (input text). I believe it will be difficult to find this algorithm, since |ideal_layout - computed_layout| could be different for every use case even if the declaration is a constant.
There are a couple ways to solve this problem. The layout output from your algorithm should instead be an ansatz. You present the "suggested" layout for a given declaration, then let the user fine-tine the positioning if they need to for their use-case.
Another way to solve the problem would be to allow the user to specify positional weights in the declaration which your layout generator uses as additional input to decide where to place nodes.