Google launches Gemini Omni 1.2 Flash for cheaper 15-second AI video

Expanding video length and reach

Google DeepMind has introduced Gemini Omni 1.2 Flash, the latest iteration of its multimodal video generation and editing model. This new version allows for the creation and modification of native video clips that reach up to 15 seconds in length. This is a notable jump from the 10-second limit found in the previous Omni 1.1 Flash version.

The model is designed to handle a variety of inputs, including text, images, audio, and existing short video clips. It produces video outputs that include synchronized audio, making it a comprehensive tool for content creators. The model is currently available for use through the Gemini API and Google Cloud platforms.

While the model offers expanded capabilities, Google notes that certain complex tasks still require further development. These include fast sports action, rapid changes in scripts, and complex physics. Despite these limitations, the model aims to provide a more seamless experience for standard video generation tasks.

Significant price reductions

One of the primary updates with the 1.2 Flash release is a substantial decrease in cost for developers and enterprises. Google has priced the model at $0.07 per second of 720p video, which is approximately 30% cheaper than the previous version. This makes it more accessible for high-volume production workflows.

To put this into perspective, a full 15-second clip at 720p resolution now costs roughly $1.05. The pricing structure remains consistent across different resolutions, providing a predictable cost model for users. For those using Google Vids, these 15-second clips can be generated at no additional cost.

As MarkTechPost reported, these savings are a key takeaway for developers looking to scale their applications. By lowering the bill per second, Google is encouraging more frequent use of the model for prototyping and final production. The goal is to provide a balance between high-quality output and economic feasibility for various business sizes.

Precision in editing and consistency

Gemini Omni 1.2 Flash introduces several technical upgrades focused on visual and auditory consistency. Users can now combine character, product, or style images with motion clips that are up to 3 seconds long. This helps ensure that faces, garments, and brand colors remain stable across different camera angles with less morphing.

The model also improves camera control, specifically for movements like FPV flythroughs, zoom-outs, and orbits. These actions now show fewer unwanted jump cuts, which often plague AI-generated video. Additionally, the dialogue sounds more natural, featuring clearer lip-syncing and better conditioning for distinct speakers in a scene.

For developers, the Interactions API handles editing through a multi-turn conversational workflow. This means you can request a change, such as swapping a background or adjusting the lighting, and the model will apply the fix to the existing clip. Because the system uses a previous_interaction_id, the prior video does not need to be re-uploaded for every edit.

The draft and upscale workflow

Google frames the use of Gemini Omni 1.2 Flash as a structured workflow rather than a single-step generation process. Users are encouraged to draft their content at lower resolutions, such as 360p or 720p, to save on costs and time. Once the desired details are finalized through follow-up edits, the best result can be upscaled to 1080p or 4K.

This method allows for more precise control over the final product without the waste of regenerating entire clips. It enables creators to iterate on specific details, like a product placement or a specific lighting effect, turn by turn. This iterative approach is designed to help users build complex scenes while maintaining strict control over the visual elements.

The model also includes a SynthID watermark on its outputs to help identify AI-generated content. While Google has not published specific benchmark scores for version 1.2, early feedback from partners has been positive. For example, invideo's VP of Engineering noted that the model sticks well to prompts and produces realistic physics and sound.

Availability and integration

Developers can begin prototyping with Gemini Omni 1.2 Flash in Google AI Studio or build production-ready applications using the Gemini API. For enterprise users who require specific service level agreements, the model is available on Google Cloud via the Agent Platform API. These options provide different levels of support and infrastructure depending on the user's needs.

The model is also being integrated into several existing Google products and services. This includes tools like Flow, Google Vids, and Google Ads, as well as the standard Gemini app. These integrations aim to bring the power of native multimodal video generation to a wider audience of non-technical users.

While the model is widely available, some features remain restricted. For instance, audio references are still allowlist-gated on Google Cloud. Despite these gates, the release marks a significant step forward in making high-quality, editable AI video a standard tool for both individual creators and large-scale commercial operations.