There are 129,864,880 books. That's ~ 2^27. There are 2^256 seed phrases. What you're proposing reduces entropy by a factor of 2^229, at least in theory, which is like reducing a seed phrase with 24 word seed phrase to one with 2.45 significant words (I.e. not counting the checksum). I'm not counting editions, but if the average book has 8 editions, that's only 3 bits more, so 2.72 words.

In practice, an attacker would need a digital library of all editions of all books. But your book is likely to come from a smaller set, e.g. it's likely in English and less than 100 years old.

This technique is called a one time pad and is similar to seedXOR.

Really clever, thanks!