[mkl]

Looks neat! I'm wondering why/how the first two results pictures (under "Values For num_steps") in https://tylerxhobbs.com/essays/2020/flow-fields have flow lines that are crossing each other. The paths follow a direction field, so that shouldn't happen except where the gradient vector vanishes, which doesn't seem to be the case there.

[nrjames]

There are many different ways to move objects through a flow field. Some people use the vectors to accelerate particles, others as instructions for instantaneous rotation, etc. The crossing lines aren't necessarily following the flow field in those images, but likely are just using vectors as an input to another set of instructions.

[mkl]

Yes, okay, but the algorithm described just before that treats the vectors as tangents to the path. The pseudocode does a complicated-seeming grid lookup instead of just evaluating the function, but apart from that it's just following the vectors as tangents.

[psytrx]

I love flow fields. They are simple, yet elegant and efficient. They are flexible, which Tyler Hobbs shows very nicely in the linked article.

A few months ago I stumbled over that article, and it made me instantly want to try them out. Here's an experiment I made for my website (WIP):

https://preview.fdumont.net/generative/flow-field

[joeberon]

Tyler Hobbs usually keeps a large distance between the exposition of the algorithm and the actual artistic example for the sake of maintaining artistic technique secrecy.