I looked at Wikipedia; Taler then Blind signature, then looked for docs on GNU Taler where I didn't immediately find any technical overview of how that works. Phind gave me a seemingly reasonable and understandable answer. (Surely sourced from somewhere.)

When we multiply the original message by rere, sign it, and then multiply by r−1r−1, the blinding factors cancel out while preserving the signature.

The success of this process relies on two critical properties:

• The blinding factor must be relatively prime to N
• The RSA keypair must satisfy the congruence relation red≡r(modN)red≡r(modN)

The magic is that you can

1. Apply a mathematical operation on your data
2. Sign that data
3. Revert/Invert the mathematical operation

and the signature remains valid.

It does sound like magic. But isn't most of cryptography like that?

There's a python example in there as well, with such a calculation. I didn't go through it though.