Light and simple REST servers can be implemented quicker with Flask - less boilerplate and less setup.

Large complex codebases often have design considerations that don't fit neatly into Django's one-model-per-view, all-models-are-backed-by-a-relational-database opinions - think apps that require immutable datastores or frequent calls to an ML pipeline. Those are often better implemented as a composition of various libraries.

The thing about organic codebases is that, you can impose a better architecture if you so wish. Frameworks often make the architectural decision for you, in which case you're out of luck if that architecture doesn't suit your application.