We are now in September and I am still not able to get vision to work in Assistant. I tried base64 encoding, the image interpretations did not seem to have anything to do with the pics. I also tried the URL approach, I tested with images from Wikipedia, that worked reliably every time. Then I tested with images hosted publicly on Google storage bucket via simple direct URLs… and it was so very on and off…unreliable. Has anyone really found a solution that works with Assistants?