Below is my workflow:
1. Customer uploads a document in our portal, which we transcribe into text.
2. We use the generate from text API of Gamma to pass the text for PPT generation
3. We do not configure any imageOptions.source or imageOptions.style
Generally seeing around 40-50% of the images being malformed, having spelling mistakes, or being completely irrelevant to the topic.
I will have to get x-ids for previous generations if thats helpful