abharga2

Thanks for the reply! The search results are a function of the images we have - we've only indexed 2M images so far. Also, the model prefers certain semantic patterns that we haven't automated yet. For example "a boy playing with a toy helicopter" gives more meaningful results. :)