If your metaballs are mostly the same size, the blurring step can happen ahead of time with a commutative convolution, stored in the texture. Then, rather than

If your metaballs are mostly the same size, the blurring step can happen ahead of time with a commutative convolution, stored in the texture. Then, rather than applying a threshold to the resulting pixel/voxel contribution, you can apply a sigmoid function (or something cheaper approximating it) and get smooth edges. You can cheat normals like you can with any other sort of height map, by measuring the differences between neighboring fragments. like so: https://rezmason.net/scourge/quick_metaballs.mp4 https://rezmason.net/scourge/quick_metaballs.mp4