Unfortunately, increasingly model providers resort to Google for webspace, which does not allow direct download links. Interactive downloads are nonce-backed and shortlived.

See example.

According to this Stackoverflow entry, they require users to make a second request based on some previous cookie.

This is very annoying, but luckily there is a Python package gdown which can do the dirtywork.

The problem now becomes: how do we integrate this, so other URLs are not affected, and our _download_impl's progress meter and file output still work?