Google's AI Evolution
Google Rolls Out Nano Banana 2 Lite and Omni Flash, Advancing AI Image and Video
Key Points
Nano Banana 2 Lite generally trades some texture and character consistency for speed, but Arena.ai Elo scores show its outputs are rated nearly as highly as the full Nano Banana models; it still has weaknesses like difficulty with small text, potential inaccuracies in infographics, and less consistent characters across iterations.
Nano Banana 2 Lite is listed as gemini-3.1-flash-lite-image and can replace gemini-2.5-flash-image; developers can access it through Google AI Studio, the Gemini API, and the Gemini Enterprise Agent Platform, while consumers can reach it via AI Mode in Search, the Gemini app, NotebookLM, Google Photos, Stitch, Google Flow, and Google Ads.
NB2 Lite is currently ranked No.5 on Arena's text-to-image leaderboard, indicating its standing among contemporary models in this benchmark space.
Omni Flash adds enterprise-friendly API access and supports video generation and conversational editing, enabling one model to take text, images, and video inputs to produce a finished clip with synced audio; this rollout moves Omni from consumer tooling toward a production-ready, unified workflow for organizations.
Google launched two new AI models on June 30: Nano Banana 2 Lite, its fastest and cheapest image generator yet, and Gemini Omni Flash, a multimodal video model now in public preview. Nano Banana 2 Lite generates 1,024-pixel images in four seconds and costs just $0.034 per 1,000 images — a fraction of what traditional vendors charge, according to The Decoder.
Together, the releases push Google from consumer AI tools toward full enterprise infrastructure. The goal is one platform that handles text, images, and video — cutting the "vendor sprawl" developers face when jumping between different AI models for each task.
Nano Banana 2 Lite Trades Polish for Raw Speed
Nano Banana 2 Lite — listed in the API as gemini-3.1-flash-lite-image — sits at No. 5 on Arena.ai's text-to-image leaderboard with an Elo score of 1,251. That edges out the full Nano Banana Pro, which scores 1,245, according to Ars Technica. The speed gain is real: four seconds per image, compared to longer waits on previous models.
The tradeoffs are real too. The model can struggle with small text and infographic data. Characters may look different from one image to the next. Product managers Alisa Fortin and Anish Nangia at Google DeepMind called it the "fastest, most cost-efficient Gemini Image model" for high-volume pipelines — but analysts at NDTV Profit note it is built for creative scale, not creative depth.
$0.034 Per 1K Images Could Upend Stock Photo Markets
At $0.034 per 1,000 images, the economics are striking. Generating 1,000 asset sets through the API now costs roughly $900. Traditional vendors charge $50,000 or more for the same volume, according to TechCrunch. That gap could hit stock photo agencies and localization firms hard.
WPP CIO Elav Horwitz said the model represents a "leap forward for controlled AI production," pointing to asset localization and product swaps as key use cases. Developers can access it through Google AI Studio and the Gemini API. Consumers get it inside Search, the Gemini app, Google Photos, Google Ads, and NotebookLM.
Omni Flash Brings Conversational Video Editing to Enterprises
Gemini Omni Flash is now in public preview via the Gemini API and the Gemini Enterprise Agent Platform. It outputs 720p video clips — ranging from 3 to 10 seconds — with native audio, at a cost of $0.10 per second of video, according to The Decoder. One model takes text, images, or existing video and produces a finished clip with synced audio.
Instead of a traditional timeline editor, users talk to the model. They can say "swap the character" or "relight the scene" and the model makes the change. Andrew Carr, Chief Scientist at Cartwheel, praised its "excellent ability to generate and edit shots that have dynamic camera motion" — a common weak spot in earlier video models.
Google Bets Ecosystem Lock-In Beats Raw Quality
Google's top competitor on image benchmarks is still GPT Image 2, which scores between 1,387 and 1,512 on the Arena.ai leaderboard — well above Nano Banana's 1,251. Ars Technica noted that Google is not yet as fast or customizable as Krea 2 Turbo. But Google's edge is not pure quality — it is integration.
Both models plug directly into Google Workspace, Google Ads, and the Gemini Enterprise Agent Platform. Legacy models including Imagen 4 are scheduled for sunset on August 17, 2026. Google DeepMind CTO Koray Kavukcuoglu framed the releases as a shift toward "coherent world simulation" — one framework for images, audio, video, and text together.
Sentiment
Mostly PositivePublishers
17
Articles
18
Reach
35
