The 1:1 case is not too unreasonable, if you ask me. Store a cryptographically signed photo on a chip in the ID card, and the machine can compare that to the real-life human presented. There are obviously drawbacks with biases in the comparison algorithm, but that's not really any worse than a human doing the same. From a privacy perspective it's not too bad - provided they delete their copy of the photo after the card is issued. This isn't any different from having the photo printed on the ID card, if you ask me.
The 1:many case, on the other hand, is a bit of a problem. This requires the creation of a mass database with everyone's pictures in it. The privacy implications are obviously enormous, as it would also enable the identification of previously-anonymous people "in the wild".
And then there's obviously the issue that it simply cannot possibly work on a larger scale: with a 1:1 comparison you have to look for a close-enough match of a single picture pair, but with a 1:many comparison you have to identify one person out of millions of possible matches. There needs to be some lenience in the matching (people use makeup and get rhinoplasties) but people's faces already look quite similar - once you get to the million-people scale, there are pretty much guaranteed to be some false positive matches in there!