A generated forest can look lovely until I need to move the path two metres to the left. If the output is one flattened image, that small revision becomes another generation with more opportunities to change things I wanted to keep.
A procedural scatter gives me different controls. I can change the path’s exclusion area, adjust density and keep the seed. A model could help propose the assets or build the initial graph, while the graph handles the arrangement.
That combination interests me more than choosing between the two approaches. Generate a starting material, then expose its useful parameters. Suggest an arrangement, then keep the relationships available for editing.
Reproducibility still needs care. Save the seed, parameters, asset versions and graph. A seed alone won’t reproduce a result if the generator or its inputs have changed.
I enjoy making decisions through direct manipulation. An AI tool can get me to a useful starting point sooner, but I’d like the result to have handles I can actually grab.