Blog

Gemini Omni 1.2 Flash: Fast and cost-effective with longer clips

Anish Nangia

2026.11.07

·Web·by Homin.Lee

#AI#API#DL#RAG#REST

Key Points

1Gemini Omni 1.2 Flash is now generally available, offering native video generation of up to 15 seconds at a 30% lower cost of $0.07 per second for 720p resolution.
2The model features enhanced reference adherence and improved character or object consistency, allowing for more precise control over multi-turn conversational edits without needing to re-generate clips from scratch.
3Advanced capabilities now include fluid multi-speaker dialogue, complex continuous camera movements, and reliable integration across Google's developer platforms and enterprise applications.

Gemini Omni 1.2 Flash, released on November 5, 2026, represents a significant advancement in generative video AI, focusing on increased duration, improved consistency, and cost-efficiency. Available via the Gemini API and Google Cloud infrastructure, the model reduces costs by 30% compared to its predecessor (Omni 1.1), priced at \0.07$ per second for 720p output.Core Capabilities and Improvements The model extends native video generation from 10 seconds to 15 seconds, allowing for one-second increment granularity (3 to 15 seconds). It introduces enhanced reference adherence, enabling users to maintain character, product, and style consistency across complex camera movements, transformations, and multi-turn conversational edits.Methodology and Technical Features Multi-turn Conversational Editing: Unlike iterative re-generation, Omni 1.2 Flash allows users to perform targeted modifications (e.g., swapping backgrounds or lighting) while preserving existing character performances, object reflections, and camera framing. This is achieved through a state-maintained architecture that supports persistent identity across multiple editing turns. Enhanced Consistency Framework: The model improves visual stability by minimizing drift and morphing between motion and camera transitions. It utilizes sophisticated interpolation techniques between keyframes, ensuring that materials, shadows, and typographic details (such as specific brand text) remain locked across extended sequences. Reference-Guided Generation: Users can condition outputs using multi-modal inputs, including static images, short motion clips (up to three seconds), and audio cues. This is applied to complex tasks such as virtual try-ons—where the model maintains garment drape and model identity—and cinematic camera movements, such as continuous FPV drone flythroughs that avoid jump cuts. Audio-Visual Synchronization: The model features improved natural speech generation, utilizing multi-speaker conditioning. By sampling voice references alongside character and mouth-shape references, the model performs expressive lip-syncing with high fidelity to comedic and dramatic timing. Scalable Workflow: The model supports an efficient production pipeline where drafts are generated at lower resolutions (360p or 720p) for rapid iteration, followed by upscaling to 1080p or 4K for final delivery. Enterprise and Developer Integration Omni 1.2 Flash is built for scale, integrated into Google Cloud with support for Provisioned Throughput. To ensure safety and provenance, all outputs are marked with an imperceptible SynthID watermark. The model is accessible to developers via the Gemini API and Google AI Studio, and to business/general users through Google Vids, Flow, and Google Ads.

Blog

Gemini Omni 1.2 Flash: Fast and cost-effective with longer clips

Anish Nangia

2026.11.07

·Web·by Homin.Lee

#AI#API#DL#RAG#REST

Key Points

1Gemini Omni 1.2 Flash is now generally available, offering native video generation of up to 15 seconds at a 30% lower cost of $0.07 per second for 720p resolution.
2The model features enhanced reference adherence and improved character or object consistency, allowing for more precise control over multi-turn conversational edits without needing to re-generate clips from scratch.
3Advanced capabilities now include fluid multi-speaker dialogue, complex continuous camera movements, and reliable integration across Google's developer platforms and enterprise applications.

Gemini Omni 1.2 Flash, released on November 5, 2026, represents a significant advancement in generative video AI, focusing on increased duration, improved consistency, and cost-efficiency. Available via the Gemini API and Google Cloud infrastructure, the model reduces costs by 30% compared to its predecessor (Omni 1.1), priced at \0.07$ per second for 720p output.Core Capabilities and Improvements The model extends native video generation from 10 seconds to 15 seconds, allowing for one-second increment granularity (3 to 15 seconds). It introduces enhanced reference adherence, enabling users to maintain character, product, and style consistency across complex camera movements, transformations, and multi-turn conversational edits.Methodology and Technical Features Multi-turn Conversational Editing: Unlike iterative re-generation, Omni 1.2 Flash allows users to perform targeted modifications (e.g., swapping backgrounds or lighting) while preserving existing character performances, object reflections, and camera framing. This is achieved through a state-maintained architecture that supports persistent identity across multiple editing turns. Enhanced Consistency Framework: The model improves visual stability by minimizing drift and morphing between motion and camera transitions. It utilizes sophisticated interpolation techniques between keyframes, ensuring that materials, shadows, and typographic details (such as specific brand text) remain locked across extended sequences. Reference-Guided Generation: Users can condition outputs using multi-modal inputs, including static images, short motion clips (up to three seconds), and audio cues. This is applied to complex tasks such as virtual try-ons—where the model maintains garment drape and model identity—and cinematic camera movements, such as continuous FPV drone flythroughs that avoid jump cuts. Audio-Visual Synchronization: The model features improved natural speech generation, utilizing multi-speaker conditioning. By sampling voice references alongside character and mouth-shape references, the model performs expressive lip-syncing with high fidelity to comedic and dramatic timing. Scalable Workflow: The model supports an efficient production pipeline where drafts are generated at lower resolutions (360p or 720p) for rapid iteration, followed by upscaling to 1080p or 4K for final delivery. Enterprise and Developer Integration Omni 1.2 Flash is built for scale, integrated into Google Cloud with support for Provisioned Throughput. To ensure safety and provenance, all outputs are marked with an imperceptible SynthID watermark. The model is accessible to developers via the Gemini API and Google AI Studio, and to business/general users through Google Vids, Flow, and Google Ads.

View original