Gemini Pro Keeps Losing Context After a Few Messages

Hi everyone. I’m struggling with Gemini Pro. We have great conversations about company marketing strategy, website planning, etc., it gives really good ideas, we get through about five questions, and then I ask it to create the agreed homepage. At that point it suddenly starts messing up, saying it’s a text-based AI that can’t generate images, even though I’ve enabled the image generation option. We go back and forth for a few questions -“just do it, here you go” – and it keeps replying with things like “I can’t do that,” “I can do it,” “actually I can’t,” etc. If I only ask it to generate images, I can open a new chat and it does exactly what I want, but then I lose all the earlier context that would be useful for the design.

I also tried Gems, but I run into a similar issue: I set up a Gem, iterate and fine-tune it maybe ten times, and it works – then two questions later it loses the thread again and I have to restart the whole Gem. Meanwhile, with ChatGPT I have conversations that have been running for weeks and it still remembers everything from the beginning.

The reply shown in the screenshot is where I finally lost my patience. Not because it failed to generate the image for the tenth time, but because the phrase at the end -“placed in a lifelike environment” -was pulled from another, older conversation where it was generating product photos. In this conversation nothing like that was ever mentioned.

It’s unbelievable to me that it loses the thread after a few questions in the same chat, especially after all the hype about its massive context window and how much text it can supposedly handle.

Has anyone figured out a solution for this so I don’t have to restart the conversation and paste in pages of text again?