The normal base model I use doesn't do this concept well, so I figured I would make a LoRA for it.

This LoRA does take a little work to get things right, but works fairly well when it does. Weights of 0.5-0.7 are the best, with higher weights being better. Basically, the higher the weight the better the placement. Lower weights will just put the stream wherever. Stream size is a bit hard to prompt for, but big things like mountains = big streams.

And for you people who like feet, it can do some of that too. Mostly by chance.

All sample images were generated at a strength of 0.7 using epiCRealism.