I seem to remember many years ago there was a search engine for images. I believe you fed it text and it would pull up a series of images. You then clicked on the image that looked most like the one you were searching for and the process would repeat until you got what you wanted.
The weird thing is, I remember this being around the late 2000s, but it doesn’t seem like the technology would have existed to do that in real time.
That depends on how fancy the algorithm was for determining how similar two images are. By the simplest algorithms, a beigeish face against a black background is very similar to a beigeish vegetable against a black background, but a beigeish face against a black background would be very different from a brown face against a white background. Nowadays, though, nobody working in image processing/recognition/etc would consider such a simplistic algorithm to be at all worth bothering with.