Gemini Omni Flash Unifies Video Generation at 10 Cents a Second

Google DeepMind released Nano Banana 2 Lite and Gemini Omni Flash to the Gemini ecosystem. The updates reduce latency for image generation and transition video generation to a unified transformer architecture.

High-Velocity Image Generation

Nano Banana 2 Lite operates under the model name gemini-3.1-flash-lite-image. It functions as a direct drop-in replacement for the previous gemini-2.5-flash-image endpoint. The architecture prioritizes execution speed and cost efficiency for production workloads over high-end creative nuance.

Outputs take approximately 4 seconds to generate. This cuts the latency of the original Nano Banana 1 model in half. The model maintains prompt adherence and improves rendering accuracy for complex text inside images.

Metric	Nano Banana 1	Nano Banana 2 Lite
Model Name	gemini-2.5-flash-image	gemini-3.1-flash-lite-image
Generation Time	~8 seconds	~4 seconds
Cost per Image	Not specified	$0.034

For developers attempting to reduce LLM API costs in production, the 3.4-cent unit cost alters the constraints around automated image generation at scale. The endpoint is generally available in Google AI Studio and the Gemini API.

Unified Architecture for Video

Gemini Omni Flash replaces Veo 3.1 as the primary video generation engine. Previous iterations relied on a split-stack pipeline that passed data between dedicated text, image, and video models. Omni Flash processes all modalities simultaneously in a single forward pass.

The model supports conversational video editing. Developers can pass natural language prompts to swap characters, adjust lighting, alter camera angles, or swap background environments. The multimodal input accepts up to five reference photos to lock visual consistency for specific characters or objects. If your system relies on cross-modal RAG pipelines, you can now feed retrieved images directly into the video generation context.

Outputs include natively synchronized audio tracks generated alongside the visual frames. Clips face a hard limit of 10 seconds per generation. Google estimates pricing at $0.10 per second of video output during the current public preview phase. All generated media includes SynthID digital watermarking and C2PA credentials.

Enterprise Availability

Both models immediately support integration with the Gemini Enterprise Agent Platform to enable visual workflows for autonomous systems. They also power Google Flow, a new creative production workspace, and serve as background generation tools for YouTube Shorts creators.

If your codebase calls gemini-2.5-flash-image, update your routing logic to gemini-3.1-flash-lite-image to capture the latency improvements. For video generation workloads, audit your pipeline constraints to account for the Omni Flash 10-second clip cap before migrating away from existing Veo configurations.