Gemini Omni 1.2 Flash release notes

Overview

Google DeepMind has launched Gemini Omni 1.2 Flash, a new iteration of its video generation model that is now generally available across the Gemini API and Google Cloud enterprise services. This release introduces longer video durations, enhanced consistency in character and object rendering, and more efficient conversational editing workflows.

Extended Video Capabilities

Gemini Omni 1.2 Flash extends the maximum native video generation length to 15 seconds, up from the 10-second limit in version 1.1. Users can specify durations in one-second increments between 3 and 15 seconds, providing greater flexibility for creating social media advertisements, product explainers, and narrative content.

Improved Consistency and Reference Adherence

The model features significant upgrades in how it maintains visual fidelity across generated frames.

  • Reference Following: The model better preserves details such as faces, garments, packaging typography, and brand colors when guided by character sheets or product photography. This enables applications like virtual try-ons and branded catalog spins.
  • Visual Stability: There is a marked reduction in visual drift and morphing during motion or camera transitions. The model is better at maintaining character, object, and scene consistency from the first frame to the last.
  • Cinematic Control: The model supports complex, continuous camera movements—such as FPV flythroughs and orbits—with fewer unwanted jump cuts and more stable scaling.

Conversational Editing and Workflow Efficiency

Gemini Omni 1.2 Flash supports multi-turn conversational editing, allowing users to refine specific elements of a clip without regenerating the entire sequence. Users can modify backgrounds, lighting, weather, or aspect ratios in subsequent turns, which facilitates the creation of ad variants for different audiences.

To optimize costs, the model supports a tiered workflow: users can draft content at lower resolutions (360p or 720p), perform edits, and then upscale final assets to 1080p or 4K. The pricing for 720p video generation has been reduced by 30% compared to the previous version, now set at $0.07 per second.

Audio and Speech Integration

The model provides improved support for natural-sounding dialogue, including clearer lip-syncing and distinct multi-speaker conditioning. These capabilities are integrated into tools like Google Vids and Flow, allowing for the creation of polished videos with synchronized narration and character dialogue.

Availability and Enterprise Features

Gemini Omni 1.2 Flash is now generally available for developers via the Gemini API and for enterprise users through Google Cloud.

  • Enterprise Security: The model is backed by Google Cloud’s enterprise-grade infrastructure and security protocols.
  • Provisioned Throughput: Enterprise users can access provisioned throughput for high-scale applications.
  • SynthID: All outputs generated by the model carry an imperceptible SynthID watermark to assist in content identification.

Sources