Except they totally could. But a data source of such size of material where everyone opted in to use for AI explicitly does not exist. The reason they fight it is in part also because training such models isn’t exactly free. The hardware that it’s done on costs hundreds of thousands of dollars and must run for long periods. You would not just do that for the funsies unless you have to. And considering the data by all means seems to be collected in legal ways, they have cause to fight such cases.

It’s a bit weird to use that as an argument to begin with since a party that knows they are at fault usually settles rather than fight on and incur more costs. It’s almost as if they don’t agree with your assertion that they needed permission, and that those imagines were IP violating 🤔