How to randomly pad files before encryption to prevent file fingerprinting?

Just use rclone. It does this natively.

Yes GPG should add appropriate padding (random initialization vector) to not reveal whether two ciphertexts have the same plaintext. It makes no real attempt to conceal that the two plaintexts have the same length. If you want that, best bet is to make all ciphertexts the same length, by padding plaintexts out to 1MB or whatever, and turning off compression. Actually you might first check the manual to see if there is already an option for that. There are a lot, and I no longer keep track.

Cryptographer's saying (Silvio Micali, I think): "A good disguise does not reveal the person's height". So you are on the right track.

Any encryption worth being called that randomises itself such that the same file, encrypted twice with the same key, produces completely different ecrypted versions.

To put it another way, if similarities in files could be detected despite their encryption, then the encryption would be worthless.

https://en.m.wikipedia.org/wiki/Ciphertext_indistinguishability

You could use Duplicati. I'm not sure if it does padding but it simplifies encryption and prevents corruption.

Sounds like you want a proper backup solution. Take a look at borg backup, a tool that supports encrypted, deduplicated, compressed, incremental backups. You can even directly save to your cloud via protocols such as ssh, s3, etc.

Have you considered cryfs?

check out PicoCrypt

If there are really no good padding tools I'll probably start a new git repo and write one, would be a pretty handy CLI tool