Google Gemini Leak Reveals the Missing Link in Image Generation

Generating AI images in Google Gemini has historically come with a major annoyance: if the result isn't perfect, you’re stuck. You can’t tweak a small detail; you have to either re-roll the dice with a new prompt or download the file and fix it in Photoshop. That looks to be changing.

From Static Images to Active Editing

Right now, Gemini acts solely as an image generator—you get what you get. The leaked update suggests a shift toward a proper editor. As identified by TestingCatalog, Google is building tools that let users draw on and add text to images without ever leaving the browser tab.

But the leak points to something more useful than just doodling on a picture. The real utility here appears to be "inpainting," or localized, prompt-based editing.

Instead of rewriting an entire prompt because the AI messed up a hand or a background detail, users could simply circle the problem area and tell Gemini, "Fix this." This aligns with discoveries on the mobile version of the app, where hidden text pop-ups suggest these markup tools are designed to guide the AI's revisions. It’s the difference between asking for a whole new image and simply telling the AI to "add sunglasses to the dog" in a specific spot.

Great Tech, Weird Name

These workflow fixes are long overdue, especially given the power under the hood. The current model—which carries the bizarre internal codename "Nano Banana Pro"—is based on the Gemini 3 Pro architecture.

While tech enthusiasts like Mishaal Rahman have praised the "Nano Banana Pro" output quality, the actual user experience has fallen flat because of the software limitations. Great image generation doesn't matter much if the editing process forces you to leave the ecosystem. By integrating these tools directly, Google would finally match the interface's utility to the model's capabilities.

Since this comes from a leak, there’s no official release date. However, the presence of these tools in the code suggests Google knows it needs to close the gap between generating an image and actually finishing it.