I am currently struggeling trying to set up a working prototype for the firebase extension: multimodel task with gemini api. It seems to connect and answer each prompt based on text but it refuses to take any kind of image. I checked every possibilty of feeding it an image. I have put the image field var into the prompt {{url/base64}} and tried to remove it for both extern images from websites that are publicly available and also an completly open storage bucket in the same firebase project. It just wont see or reconize the picture and i dont get any errors. If specificly asked to analyse the given image it just halucinates anything random that has surely nothing to do with the real image.

I am out of any ideas to fix this problem already installed deinstalled the extension relabeled any field for the request document used any kind of image in any size and tryed to change the model although i thing gemini vision pro 1.0 is the only one i can acces.

Does anyone know of such a problem and what to do to fix it?

Thanks a lot!

See the first paragraph i am new to this and i didnt know about this structure