Python: How do I test an image generation tool?

I've create a small tool to generate images (from other images) using PIL. I'm now wondering how I can create automated tests for this as the test would need to compare the generated image to a template image (which was also generated with this tool)?

I expect if I make changes or one of the underlying libraries make changes the images won't be identical on the bit level even if they look identical to humans. What is the common practice in such scenarios? Use a similarity measure and just fail the test below a certain threshold?