Google launches Gemini Omni 1.2 Flash for longer video generation

Expanded video length and scale

Google has officially rolled out Gemini Omni 1.2 Flash across its Gemini API and Gemini for business platforms. This new model iteration allows for the native creation of video clips reaching 15 seconds in length. Users can specify the exact duration in one-second increments ranging from 3 to 15 seconds.

The update is designed to help creators produce more complete content in a single generation. This includes items like product explainers, social media advertisements, or short storytelling sequences. By increasing the limit from the previous 10-second cap, the model provides more room for narrative flow.

To keep costs manageable for high-volume production, Google has lowered the price by 30% compared to the previous version. The new rate is $0.07 per second for 720p video. This pricing structure is intended to make the tool more viable for large-scale commercial applications.

Consistent brand and character details

One of the primary improvements in Gemini Omni 1.2 Flash is its ability to adhere to specific references. The model can process a combination of character images, product shots, style images, and short audio cues. It is designed to keep details like faces, clothing, and brand colors consistent across various camera angles.

This capability is particularly useful for creating virtual try-on sequences or catalog spins. For example, the model can take a photo of a person and three different outfits to generate a 15-second video showing 360-degree turns. It maintains the specific cut, drape, and back details of each garment throughout the motion.

The model also handles complex typography and materials with higher precision. In testing, it was shown to build posters layer by layer while locking in specific materials like ribbed glass or coiled-cord letterforms. This reduces the visual drift or morphing that often occurs during camera transitions or animations.

Conversational editing and workflows

The new model introduces multi-turn conversational editing, which allows users to refine a clip turn by turn. Instead of starting a generation from scratch when a mistake is made, users can request specific changes. This includes swapping out a background, adjusting the weather, or changing the lighting of a scene.

This workflow is built to save time and money for professional creators. A user can generate a base clip and then iterate on it to create multiple variants for different audiences. For instance, a character's performance and armor reflections can remain the same while the background shifts from a forest to a castle.

Google also highlights a tiered production workflow to optimize costs. Creators can draft scenes quickly at 360p or 720p to find the right look. Once they are satisfied with a specific detail, they can use a follow-up edit to fix it before upscaling the final result to 1080p or 4K resolution.

Cinematic motion and natural speech

Gemini Omni 1.2 Flash aims for a more polished, cinematic look by handling complex camera movements. It can execute sweeping FPV flythroughs or dramatic zoom-outs with fewer unwanted jump cuts. The model is designed to maintain a steadier scale during these continuous movements to ensure a smoother visual experience.

The model also improves the quality of dialogue and lip-syncing within generated scenes. It can route different sampled voices to the correct characters based on provided voice references. This allows for more natural comedic timing and expressive lip-syncing in animated scenes or video explainers.

These features are being integrated into several Google products, including Flow and Google Vids. For enterprise users, the model is backed by Google Cloud's infrastructure and includes an imperceptible SynthID watermark. This watermark helps identify the content as AI-generated while remaining invisible to the human eye.

Availability and next steps

The model is currently rolling out across various platforms for different types of users. Developers can prototype with the model in Google AI Studio or explore the Gemini API documentation. Businesses can access the tool through the Google Cloud console to build custom enterprise solutions.

General users can also access these features through existing Google apps. This includes generating 15-second clips in Google Vids with a Workspace subscription or using the Gemini app. Google Ads users can also begin scaling their video campaigns using the new model's capabilities.

While the model is now generally available, Google notes that development is ongoing. Future improvements will focus on areas like fast sports action, rapid script changes, and complex physics. For now, the model is positioned as a tool for creators and businesses producing video content at scale.