Google releases Gemini Omni 1.2 Flash for cheaper 15-second video generation

Expanding video length and reach

Google is rolling out Gemini Omni 1.2 Flash, making it generally available across the Gemini API and Gemini for business. This new iteration allows for the native creation of video clips up to 15 seconds long. This is an improvement over the 10-second limit found in the previous Omni 1.1 version.

Users can now specify the duration of their clips in one-second increments starting from 3 seconds. This flexibility allows creators to produce complete stories, product explainers, or full social media advertisements in a single generation. The model is designed to handle these longer sequences while maintaining visual stability.

The release also includes a significant price reduction for developers and enterprises. Google dropped the cost by 30% from the previous version, setting the price at $0.07 per second for 720p video. This move aims to make high-quality video generation more cost-effective for scaling production.

Consistent details and references

One of the primary improvements in Gemini Omni 1.2 Flash is its ability to hold onto specific details from start to finish. It can combine character, product, or style images with short motion clips and audio cues to guide the generation. This ensures that faces, garments, and brand colors stay consistent across various camera angles.

The model is particularly useful for creating virtual try-on sequences or catalog spins. For example, it can take a model photo and three different outfits to generate a 15-second sequence. It maintains the identity of the model and the specific cut and drape of each garment throughout 360-degree turns and match cuts.

The system also reduces visual drift or morphing during complex transitions. Whether a user is interpolating between keyframes or extending a clip, the products and environments stay locked in place. This is a notable upgrade over previous versions, which struggled to match reference images as closely during multi-layer animations.

Conversational editing workflows

Gemini Omni 1.2 Flash introduces a multi-turn conversational editing capability that changes how users interact with AI video. Instead of re-generating a clip from scratch when a detail is wrong, users can provide follow-up edits. This allows for swapping products, changing backgrounds, or adjusting lighting through simple prompts.

The workflow is designed to keep costs down by encouraging a draft-and-refine approach. Users can draft fast at lower resolutions like 360p or 720p and then upscale their best creations to 1080p or 4K. This method avoids the need to pay for multiple full-scale generations of the same scene.

This capability is particularly useful for creating multiple ad variants for different audiences. A user can start with one clip and use conversational prompts to swap a pavilion tent for an oak forest or a stone castle. The characters' performances and camera framing remain intact while the environment changes turn by turn.

Cinematic motion and speech

The model offers more controllable and cinematic camera movements, such as sweeping FPV flythroughs or dramatic zoom-outs. It achieves these complex motions with fewer unwanted jump cuts and steadier scaling. This results in a more polished look compared to earlier iterations of the technology.

Audio and speech capabilities have also been enhanced to produce more fluid and natural-sounding dialogue. The model supports clearer lip-sync and distinct multi-speaker conditioning. This means it can route different sampled voices to the correct characters in an animated scene with expressive timing.

These features are being integrated into several Google products, including Flow and Google Vids. For example, users can create video explainers or polished pro videos with avatars and narration. The model also includes an imperceptible SynthID watermark to identify the AI-generated content.

Availability for businesses

Gemini Omni is now generally available and backed by Google Cloud’s enterprise-grade infrastructure and security. This means businesses can start building enterprise solutions using the API with Provisioned Throughput options. It is designed for developers and creators who need to produce video content at scale.

Developers can begin prototyping in Google AI Studio or explore interactive starter apps via the developer documentation. For businesses, the model is accessible through the Google Cloud console. These platforms provide the necessary tools to integrate the model into existing commercial workflows.

General users can also access the technology through various Google apps. This includes creating 15-second clips in the Gemini app for Ultra subscribers or using it in Google Ads. The rollout ensures that both individual creators and global brands have different ways to interact with the new video generation tools.

Next steps for development

While the current version is ready for production, Google is still working on further improvements. Development is ongoing to enhance the model's performance in specific areas like fast sports action and complex physics. They are also looking to improve how the model handles rapid script changes during the generation process.

The current focus remains on providing a reliable tool for businesses and developers who need consistent results. By prioritizing reference adherence and conversational editing, Google aims to make AI video a practical tool for marketing and content creation. The model serves as a foundation for more complex video production tasks.

As the technology matures, users can expect more refined interactions and even longer sequences. The current 15-second limit provides a starting point for social ads and product explainers. For now, the focus is on providing a fast, cost-effective way to turn static images and prompts into high-quality motion.