as I've understood it you've trained a tensorflow model to recognize images of snakes and spiders, and then made a chrome extension that blocks any image the model recognizes from being displayed?

If you allowed people to create models to block by feeding in a google image search, or perhaps other open image feed then they could write in a text that when searched on Google images that returns stuff they want to block, new blocker trains on this model and they turn on that blocker.

Step 3 - allow them to run specific blockers on specific sites.