Scalability Benchmarking and combining text and image

[Better-Boy]

I have 3 questions:

1. How will I combine Text and Image instead of text to text, image to image...? Example: Consider there is a product which has images as well as textual information, how will I use GNES to get the most similar matching product from the index?

2. How scalable is this solution? Can I index 100 million images or text?

3. What to do if I want to save the indexes to database and load them during run time?

[JohnGiorgi]

Also very interested in the scaling question...

In particular, if you wanted to index an entire scientific literature dump (say, 100s of millions of papers), is this feasible?