Google Launches Gemini Omni Flash and Nano Banana 2 Lite to Power High-Speed Image-to-Video Pipelines

A High-Speed Creative Pipeline Emerges

Google released two creative artificial intelligence models on June 30, 2026, designed to work together as a rapid image-to-video assembly line. Released through the Gemini API, Google AI Studio, and the Gemini Enterprise Agent Platform, these models aim to lower the cost and latency barriers for high-volume creative production. By combining Nano Banana 2 Lite for rapid image generation with Gemini Omni Flash for conversational video editing, developers and marketing teams can instantly generate a static visual and animate it. This setup targets the expensive and slow aspects of digital marketing, such as creating dozens of localized ad variations or hundreds of product mockups.

Nano Banana 2 Lite: Speed and Cost Efficiency

The new image model, officially named Gemini 3.1 Flash-Lite Image and accessible via the API ID gemini-3.1-flash-lite-image, is the fastest and most cost-effective entry in Google's four-tier image family. This family includes the legacy Nano Banana model, the standard Nano Banana 2, the high-end Nano Banana Pro, and the new speed-focused Lite version. Generating a 1K-resolution image in approximately four seconds, Nano Banana 2 Lite costs 0.0336 dollars per one thousand images under standard pricing, which drops to 0.0168 dollars per one thousand images for batch requests. Input costs are set at 0.25 dollars per million tokens for text, image, or video inputs.

Google built the model with improved world knowledge, typography capabilities for rendering legible text directly in ad layouts, and character consistency across sequential generations. Early adopters are already integrating the model into production. Manus AI uses Nano Banana 2 Lite for autonomous workflows like slide decks and web pages, while creative leads at Artlist and Figma praise its ability to keep creators in the flow without waiting on progress bars. The model is also rolling out across Google consumer and business surfaces, including Google Flow, Google Ads, NotebookLM, Stitch, and AI Mode in Search.

Gemini Omni Flash: Conversational Video Editing

Operating under the API identifier gemini-omni-flash-preview, Gemini Omni Flash is a video generation model currently available in public preview. Priced at 0.10 dollars per second of generated video, Omni Flash allows creators to edit footage using natural language rather than a traditional editing timeline. Users can describe changes in plain text, and the model executes edits such as swapping objects, transferring styles, or relighting scenes while maintaining the original audio.

The model accepts multimodal inputs, combining text, reference images, and existing video clips. A key feature is its native audio generation, meaning sound effects are produced directly from visual events rather than being pulled from a pre-recorded sound library. Google showcased this capability through its Omni Product Studio demo, which transforms static product photographs into cinematic video advertisements.

Real-World Performance and Sandbox Testing

Independent testing of Gemini Omni Flash reveals both its breakthrough capabilities and its current limitations as a preview model. When running the model inside Gemini with Flash 3.5 on Google AI Plus or Pro, tests show that Omni Flash can maintain a conversational editing chain of up to five sequential edits before its scene memory begins to break down.

In physical simulation tests, such as a marble rolling down a ramp and hitting a cymbal, the model successfully demonstrates realistic momentum, decelerating on uphill climbs and accelerating on drops. The generated audio, such as the metallic ring of the cymbal, aligns precisely with the visual impact. However, consistency remains variable. In some generation attempts, objects like marbles teleported back to previous positions or launched along physically impossible trajectories. Additionally, early testers note that the model handles object swaps much more cleanly than full background swaps, indicating that while the image-to-video pipeline shows immense promise, it still operates as a generative lottery in complex scenarios.

Whether enterprise teams will tolerate the occasional physics-defying glitch in exchange for a massive drop in production costs remains the critical question as Google pushes these rapid-fire models into live ad-tech pipelines.