Google releases Gemini Omni 1.2 Flash with 15-second video generation

The updated model introduces longer clip durations, improved reference adherence, and a 30% price reduction for developers.

Image: Google DeepMind

Google has launched Gemini Omni 1.2 Flash, an update to its generative video model that extends native clip generation to 15 seconds. This release, which is now generally available across the Gemini API and Google Cloud, aims to provide more flexibility for professional projects by allowing users to specify durations in one-second increments between three and 15 seconds.

A primary focus of the 1.2 update is improved reference adherence. The model is designed to maintain consistency for characters, products, and styles by utilizing image inputs, short motion clips, and audio cues. Google states that this allows for more stable results across camera angles, helping to prevent visual drift or morphing when generating content like virtual try-on sequences or catalog spins.

The update also introduces multi-turn conversational editing, enabling users to refine existing clips through follow-up prompts. Instead of regenerating an entire video, users can modify specific elements such as backgrounds, lighting, or aspect ratios while preserving the original character performances and camera framing. This workflow is intended to facilitate the creation of ad variants for different placements.

In addition to these functional improvements, Google has reduced the pricing for the model by 30% compared to the previous Gemini Omni 1.1 Flash. The service is now priced at $0.07 per second for 720p video generation. The company suggests a workflow where users draft content at lower resolutions before upscaling final selections to 1080p or 4K.

The model also features enhancements for audio and camera control. It supports more fluid dialogue with improved lip-sync and multi-speaker conditioning, which is integrated into tools like Google Vids and Gemini for business. Furthermore, the model is designed to handle complex, continuous camera movements, such as FPV drone flights, with fewer jump cuts and more stable scaling.

Sources

  1. Google DeepMindGemini Omni 1.2 Flash: Fast and cost-effective with longer clips
  2. Google AIGemini Omni 1.2 Flash: Fast and cost-effective with longer clips