Overview
While experimenting with Gemini’s image generation, I noticed a small but recurring friction: once an image is generated, there’s no easy way to reference it in subsequent prompts. Users who want to iterate on or modify previous images must re-enter context manually—a repeated cognitive load that slows down creative exploration.
Inspired by messaging apps like WhatsApp, I explored a concept that lets users “reply with a previous image” directly in the chat. This allows for smooth iterations without losing context, making image generation more iterative, intuitive and conversational.
Referencing Previous Images
Using Stitch, I recreated Gemini’s existing UI screens by providing the AI with prompts and attaching current screenshots. I then iteratively guided the AI to implement my proposed solution, introducing a “reply” feature to enhance the workflow.
The proposed interaction allows a user to:
Tap or select a previous image from the chat
Include it in a new prompt as a reference
Modify the prompt or experiment with variations
This keeps the workflow fluid and familiar, leveraging an existing mental model to improve usability without introducing heavy UI changes.
