Start building with Nano Banana 2 Lite and Gemini Omni Flash

Google DeepMind dropped two new models today, and the pitch is speed over spectacle. Nano Banana 2 Lite, officially gemini-3.1-flash-lite-image, generates a 1K image in about 4 seconds for roughly 3.4 cents a pop. That's the cheapest, fastest entry point yet in the Nano Banana lineup, and Google is telling anyone still running the original 2.5 Flash Image model to just swap it out — same interface, better quality, lower cost, no real downside mentioned.

The more interesting release is Gemini Omni Flash, which finally lands in the Gemini API and AI Studio after debuting at Google I/O. It handles video generation and conversational editing in the same breath, meaning you can describe a change in plain English and it reshapes the clip rather than forcing you into timeline software. Pricing sits at 10 cents per second of output, matching Veo 3.1 Fast, though right now you're capped at 10-second clips and audio-reference uploads aren't supported yet. Character consistency across scene changes still has rough edges too, which Google admits outright rather than glossing over.

What Google really wants developers to notice is the combo play. Generate a still with Nano Banana 2 Lite, then feed it into Omni Flash as a reference frame to animate it — image to motion in two API calls instead of a whole production pipeline. The Interactions API lets that workflow persist across up to three sequential edits, so a session can hold context while a user tweaks lighting, then swaps a background, then asks for movement, without starting over each time.

Google backed the announcement with three remixable demo apps: Anywhere, which teleports a selfie to landmarks and then animates the result; Space Lift, an interior design tool that turns a room photo into styled concepts and a cinematic walkthrough; and Omni Product Studio, aimed at e-commerce teams who want static product shots turned into short promotional clips. All three exist mainly to show the image-to-video handoff working end to end, not as polished products themselves.

Both models carry SynthID watermarking, and Google points to its Gemini app, Chrome integration and Search verification tools as the mechanism for checking whether content was AI-made or edited. That's become the standard disclaimer with every generative media release now, useful but not really a constraint on how the tools get used.