Knot Generation

I wanted to train a model to draw the splines of knots, both with diffusion and with GPT-style token prediction. Each knot is a path of points in 3D, so the model either denoises the whole path at once or predicts it one point at a time.

A generated knot plotted as points connected into a loop in 3D — A generated knot. Points predicted in 3D and connected into a spline.

There is a short clip of it running on YouTube.

Code: github.com/isaac-art/knot_t.