Google Pushes Gemini Omni 1.2 Flash to 15-Second Videos at 30% Lower Cost

Google has moved Gemini Omni 1.2 Flash into general availability across the Gemini API and its business products. The release generates clips up to 15 seconds, supports conversational editing, follows reference media more closely, and cuts the stated 720p generation price by 30%.

The Omni family accepts text, images, audio, and video within one prompt, then uses those inputs to generate video. General availability moves version 1.2 beyond preview for supported production workloads. Its predecessor produced clips up to 10 seconds, a limit that often required developers to stitch scenes together and manage continuity between generations.

Fifteen seconds, frame by frame

Edits reshape the cost curve

Google recommends generating drafts at 360p or 720p, correcting details through follow-up instructions, and upscaling selected clips to 1080p or 4K. Multi-turn editing keeps the prior clip in context, allowing a developer to request a new background, altered lighting, or another focused revision without rebuilding the prompt and reference setup.

The quoted rate puts a three-second 720p generation at $0.21 and a 15-second generation at $1.05. Fewer full-length retries could lower the cost of producing variants, although project budgets still need to account for edit requests, rejected generations, upscaling, storage, and downstream processing.

References carry farther

Omni 1.2 can combine character, product, or style images with audio cues and motion clips up to three seconds long. Google says the model retains those details more consistently throughout the generated scene, including during camera movement. Potential uses include product demonstrations, branded campaign variants, and virtual try-on clips in which a person’s identity and a garment’s shape must survive a 360-degree turn.

Improved reference adherence also helps with recurring characters and branded packaging. Developers should still validate logos, labels, colors, and identity-sensitive details before publishing, especially when a clip contains small or moving text.

Motion and text remain weak spots

Google continues to identify fast sports action, abrupt script changes, and complex physical interactions as difficult cases. The DeepMind model card also notes that edits can introduce continuity errors, while intricate motion and precisely rendered text may require several attempts or manual correction.

The typography claims require particular care. Omni 1.2 may preserve supplied packaging and brand references more reliably, but that improvement does not guarantee accurate spelling or stable text in every frame.

Several doors into the model

• API developers: Prototype in Google AI Studio or integrate through the Gemini API . The Interactions API maintains context across successive editing requests.
• Google Cloud teams: Access the model through Vertex AI and Gemini for business. Provisioned Throughput offers reserved capacity for workloads that need predictable availability.
• Creators: Generate and revise scenes in Flow, create 15-second clips in Google Vids, or use the Gemini app as access rolls out to Google AI Ultra subscribers.
• Advertising teams: Produce campaign assets through Google Ads.

Why the extra five seconds count

A 15-second runtime matches a common format for social advertisements, product demonstrations, and short explainers. Generating that duration as one clip reduces the need for stitching, which can expose changes in lighting, character appearance, motion, and camera position between separate outputs.

For developers, the practical gains come from combining the longer timeline with iterative editing and lower 720p pricing. Teams evaluating the model for production should test reference consistency, text accuracy, latency, regional availability, quotas, safety policies, and the full cost of revisions against their own workloads.