You want to run a transaction in production and you're unsure about the impact because it might take a lock and block other transactions for some time. In that case you may want to spin up a separate database with fake data and test it there first.

You could create a new database and restore a production backup instead of using fake data but that might not be allowed or require some kind of approval due to rules protecting the privacy of customers or employees.

Generating test data batches like this is a common use.

The other is for populating "numbers tables" which can be very useful in reporting contexts and/or dealing with certain gaps/islands type problems.

You can even use the techniques dynamically rather than a stored numbers table, if you need it in a DB that doesn't have one and you don't have schema permissions to create one, though this is generally less efficient.

Creating realistic fake data is useful in lower environments and for load testing. Outside of SQL I like faker: https://github.com/joke2k/faker

Exactly like that, yes.