Using a pre-trained model to start labeling a dataset

I have a completely un-sanitized dataset of dozens of thousands of pictures of windows, where I would like to find birds.

The idea is to take a pre-trained model, do inference on these images, and when it finds some, move to a folder, move to another folder otherwise. This would be my first iteration of this dataset, so it’s a kind of unsupervised cleanup.

Does it make sense? Is it possible?