I am transforming a mesh and doing inverse transforms later. In the end I expect geometry to be exactly the same as before transformation.

GN tree:

enter image description here

What bothers me in the first place is that GN setup is not precise enough, if compare distances between original and modified positions, one can see some unwanted offset.

Second thing I don't understand is why inverse transforms of my custom GN group via Transform node approach are less precise than via Set Position.

Here is offset comparison (left - Set Position approach, right - Transform node): enter image description here

Inside inversing node group:

enter image description here

  • $\begingroup$ If I'm interpreting your pictures correctly, you're getting errors in position of up to 1.22 E-7. That is a small number, equivalent to hundreds of nanometers at typical Blender scales, a measure that might be a useful way to measure the radius of wool fibers. As for differences in the size of this error between different techniques, look at your operations-- they are different, each operation involves floating point error, and the error accumulates after every operation. At this level of precision, rotating three times is not the same as rotating once. $\endgroup$
    – Nathan
    Sep 19, 2022 at 17:30
  • $\begingroup$ @Nathan, thanks for comment. You are right about size of error value, but this is very simple example just to point a problem. In my real project this nano-jittering easily accumulates to a point, where it messes up surface point distribution during animation. $\endgroup$
    – Serge L
    Sep 19, 2022 at 18:13


You must log in to answer this question.

Browse other questions tagged .