Gemini Imagen 3 – Conversations With AI

Of all the variations of generative AI that I’ve used, image generation seems to be the most frustrating.

On September 28, 2024, I asked Google Gemini to design a logo for my newsletter. In my experience, my request would have been a simple tasks for a human graphic designer, but it was a struggle for Imagen 3.

Even after several attempts at correction, I was unable to get the result I was looking for: a new version of my logo that overlayed the newsletter title and integrated that concept into the existing logo.

I suspected there may be a constraint that prevents the tool from copying an image too closely, so I tried another request to find out.

The results just kept getting farther away from my initial request.

I’m wondering why Google even released this version, which admits that it is “still learning.” It doesn’t seem very useful to me – at least not yet.