Google drops Nano Banana 2 Lite and Gemini Omni Flash for high-speed media creation

Drafting AI images and videos usually comes with a painful choice between waiting on rendering bars or watching your cloud credits evaporate.

On Tuesday, Google released two new generative media tools built for speed and budget: Nano Banana 2 Lite, a lightning-fast image model, and Gemini Omni Flash, a conversational video generation and editing model. Both are live today in Google AI Studio, the Gemini API, and the Gemini Enterprise Agent Platform.

Sub-cent images at 4-second latency

Nano Banana 2 Lite (internally gemini-3.1-flash-lite-image) is aimed squarely at high-throughput pipelines where shaving latency matters more than pixel perfection.

The cost equation is wild: it is priced at just $0.034 per 1,000 images, and text-to-image latency clocks in at roughly four seconds. Google pitches it as an immediate drop-in replacement for its legacy first-generation Nano Banana.

Despite the budget tag, Google claims the model preserves reliable prompt adherence, decent character consistency, and legible in-image text rendering.

Beyond developer APIs, the model is rolling out across consumer surfaces including Google Search's AI Mode, the Gemini mobile app, NotebookLM, Google Photos, Stitch, Google Flow, and Google Ads.

Chatting with your video editor

Gemini Omni Flash, previously teased at Google I/O, is also getting its public preview debut via the API and AI Studio.

The pitch here is combining multimodal reasoning with video editing. Priced at $0.10 per second of output (matching Veo 3.1 Fast), it accepts text, image, and video prompts, letting creators direct tweaks through plain conversation. In one demo, a creator performed digital magic tricks, pulling 3D text right out of a smartphone screen.

It can also pair directly with Nano Banana 2 Lite. Developers can churn through rapid image iterations with Lite, pipe the best pick into Omni Flash to animate it, and chain up to three sequential conversational edits using Google's Interactions API.

There are clear preview guardrails, though. Videos cap out at 10 seconds, scene extension and audio references are not yet supported, and consistency during pans or scene shifts can still wobble.

Both models bake in Google's SynthID invisible watermarking for content verification.

Why it matters

For builders and product teams, this drops the prototyping floor to near zero.

If you are building dynamic e-commerce pipelines, social assets, or creative prototyping boards, you can now generate thousands of rough concepts for pennies, then commit paid video compute only to the winners. It is an immediate upgrade for high-volume automated asset workflows.

Strategically, Google is carving up the multimodal market by utility rather than chasing a one-size-fits-all flagship. With Banana Pro handling heavy creative lifting, Banana 2 as the all-rounder, and Banana 2 Lite anchoring real-time apps, Google gives enterprise devs a clear pricing ladder to lock their stacks into Google Cloud. TechCrunch also noted Google recently struck a $75 million deal with indie studio A24, underlining how aggressively the company is courting the creative sector despite pushback against low-effort AI media.

We think the speed-first posture is smart, but the rough edges are obvious. Omni Flash’s character drifts and missing audio ingestion mean it cannot replace standard post-production tools yet. The risk for Google is that near-free generation might flood feeds with disposable promotional filler before the model gains enough fine control to win over serious creators.