Google releases two new generative AI models. Nano Banana 2 Lite generates images in 4 seconds at a fraction of the cost. For the first time, Gemini Omni Flash allows you to generate and edit videos with text prompts through an API.
Nano Banana 2 Lite generates images in 4 seconds
Google says Nano Banana 2 Lite is built for rapid ideation and high-throughput developer pipelines. Text to image generation takes 4 seconds and costs just $0.034 per image at 1K resolution. The new image model is gemini-3.1-flash-lite-image in the API.
| model | Price per image | solution |
|---|---|---|
| nano banana 2 light | $0.034 | 1K |
| nano banana 2 | $0.067 | 1K |
| nano banana pro | $0.134 | 1K or 2K |
Google says that despite its emphasis on speed, Nano Banana 2 Lite still offers reliable prompt following, consistent character rendering, and easy-to-read text in generated images. This model is being rolled out across Google’s consumer products, including the developer platform as well as AI mode in Google Search, Gemini apps, NotebookLM, Google Photos, Stitch, Google Flow, and Google Ads.
Nano Banana 2 Lite brings three product models to the Nano Banana family. Google positions Nano Banana 2 (Gemini 3.1 Flash Image) as an all-rounder with the best balance between quality and cost. Nano Banana Pro (Gemini 3(.1) Pro Image) is aimed at complex, professional use cases and offers what Google says is the strongest control and most advanced reasoning. 
Comparison chart for the Nano Banana model family. |Image: Google Developers can choose the right model based on whether they need speed, quality, or low cost. Google considers the original Nano Banana (Gemini 2.5 Flash Image) to be obsolete. We still primarily use Nano Banana Pro, as its image quality and quick reliability tend to outperform both Nano Banana 2 and OpenAI’s GPT-Image-2.
Gemini Omni Flash brings video generation to the API
Gemini Omni Flash was first unveiled at Google I/O and is now available to developers through the Gemini API and Google AI Studio. This model combines Gemini’s multimodal inference with video generation and editing. Pricing is $0.10 per second of video output, comparable to Veo 3.1 Fast.
Google says the model’s strengths include natural language conversational video editing, the ability to mix input formats such as text, images, and video, and the ability to leverage Gemini’s world knowledge to generate it. Text and graphics can be synced directly with video actions.
Gemini Omni Flash currently only produces 10 second clips. Audio references and scene extensions are not yet supported in the API. The API schema accepts video references up to 3 seconds in length, but according to Google, this model does not yet handle video references correctly. Character consistency across scene changes and camera movements is also still limited.
Google recommends chaining both models
Google believes that combining the two models will yield the greatest impact. Developers can use Nano Banana 2 Lite to quickly generate images, pass them as references to Gemini Omni Flash, and animate them into videos. The Interactions API, which is now Google’s default AI API, stores session history and context and can be edited up to three times in a row.
Google provides three demo apps that demonstrate how the models work together. “Anywhere” places users at famous landmarks via a selfie and animates the results. “Space Lift” generates interior design concepts from room photos and converts them into videos. “Omni Product Studio” converts static product images into EC videos.
Google says both models use SynthID watermarks to tag AI-generated content. Verification can be done through the Gemini app, Gemini in Chrome, or Google Search. Nano Banana 2 Lite and Gemini Omni Flash are now available in Google AI Studio, Gemini API, and Gemini Enterprise Agent Platform.
AI News Without the Hype – Curated by Humans
Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, exclusive six-a-year “AI Radar” frontier reports, full archive access, and access to the comments section.
Subscribe now
