Image to point cloud
![]()
A slowly shifting noise image becomes a spiky, colored form in space. Each image sample becomes a point, so no surface needs to be modelled.
RGB Simplex Noise supplies the image, and a compute shader converts it into geometry for native Model Display. Depth scaling controls how far points extend from the image plane; the threshold changes which samples contribute to the cloud.
Try it
Start playback and bring zScale toward zero to flatten the result, then increase it to recover depth. The Z component of pointScale also affects depth. Raise threshold gradually and watch which parts of the cloud remain.
No image file is needed. To adapt the patch, replace the noise source with another texture-producing process. Use a compute-capable graphics backend; the saved vertex count is about 6.4 million, so lower that workload if necessary before trying high-resolution sources.