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

Expanding video length and quality

Google has officially introduced Gemini Omni 1.2 Flash, a new model designed to push the boundaries of AI-generated video content. This latest iteration serves as an upgrade to the previous Gemini Omni 1.1 model, which was restricted to shorter clips. The new version allows users to generate videos reaching up to 15 seconds in length.

One of the primary improvements is the level of control users have over their final output. Creators can now adjust the duration of their videos in one-second increments, allowing for precise timing. This flexibility is a significant step forward for those who need specific pacing for social media or advertisements.

In addition to longer clips, the model supports high-resolution output including 1080p and 4K. Google designed the workflow so that users can begin by creating lower-resolution drafts at 360p or 720p. Once the user is satisfied with the motion and composition, they can upscale the final creation to a higher quality.

Lower costs for enterprise use

A major focus of this release is making AI video production more affordable for large-scale operations. Google has reduced the price of the Omni 1.1 Flash model by 30 percent. This move is intended to help businesses integrate AI video into their existing marketing pipelines without prohibitive expenses.

To put the pricing into perspective, generating a video at 720p resolution currently costs $0.07 per second. For a full 15-second video, the total cost would be $1.05, provided the entire clip is billed at that standard rate. These figures make it more viable for companies to experiment with high-volume content creation.

The model is being distributed through several major channels to ensure wide accessibility. It is available via the Gemini API and Google Cloud’s enterprise platform. Furthermore, Google is integrating the model into consumer and business tools like the Gemini app, Flow Google Vids, and Google Ads.

Reference-based generation tools

For professional creators, the model introduces sophisticated reference-based generation capabilities. This allows users to maintain consistency by providing the AI with specific visual anchors. Users can input character images, product photos, and even short motion clips of up to three seconds to guide the output.

These tools are particularly beneficial for commercial applications where brand identity is paramount. A business can use a specific product photo to ensure the AI renders the item accurately in a promotional scene. This reduces the randomness often associated with text-to-video prompts.

The model also incorporates audio cues to synchronize the visual elements with sound. By combining these visual references with audio inputs, creators can build more cohesive narratives. This multi-modal approach helps in producing content that feels more intentional and less like a generic AI generation.

Improved speech and lip-syncing

Google has also enhanced the way the model handles dialogue and character interaction. The updated AI features improved speech generation and better lip-syncing for the characters on screen. These refinements are designed to make the spoken parts of a video look and sound more natural to the viewer.

The model is also capable of handling multi-speaker dialogue within a single clip. This makes it a strong candidate for creating explainer videos or scenes featuring multiple characters interacting. By improving the flow of conversation, the AI can handle more complex storytelling requirements.

These technical improvements aim to solve common issues where AI-generated speech feels disconnected from facial movements. By tightening the link between audio and visual data, Google hopes to create more immersive experiences. This is especially useful for educational content where clarity of speech is essential.

Safety and watermarking

As AI-generated media becomes more common, Google is implementing measures to ensure transparency. The company confirmed that the video generation model includes an imperceptible SynthID watermark. This technology is designed to identify whether a video was created using AI tools.

A watermark is a digital signature that is not visible to the human eye but can be detected by software. This helps in distinguishing between real-world footage and synthetic media. It serves as a layer of security for platforms that want to verify the origin of the content they host.

By including this technology, Google aims to balance innovation with responsibility. It allows creators to use the tool for creative purposes while providing a way to track the source of the media. This is a key component of the company's broader strategy to manage the risks of synthetic content.