Google launches Gemini Omni 1.2 Flash with 15-second video generation and token rates

Key facts

Release context and capability milestones

On Thursday 5 November 2026 at 17:00 UTC, Google and Google DeepMind released Gemini Omni 1.2 Flash, the second generation of their multimodal video synthesis and editing model. The release establishes two primary technical milestones: expanding maximum continuous video generation from 10 seconds to 15 seconds, representing a 50.0 percent duration expansion (derived), and lowering the base video output token rate from $17.50 to $12.00 per million tokens, a 31.4 percent price reduction (derived).

Beyond continuous generation, Omni 1.2 Flash transitions video production from single-turn prompt execution into a stateful conversational workflow. In conventional video diffusion systems, modifying an isolated asset or camera angle requires re-executing the entire diffusion pipeline from noise latents, consuming full compute and introducing severe visual drift. Omni 1.2 Flash supports multi-turn conversational edits across 15-second sequences, allowing developers to preserve subject geometry, brand colors, and motion trajectories while adjusting lighting or backgrounds.

Zoom

Architectural parameters across Gemini Omni Flash model generations

Parameter	Omni 1.1 Flash	Omni 1.2 Flash	Generational Delta (derived)	Production Impact
Maximum Clip Length	10 seconds	15 seconds	+5.0 seconds (+50.0%)	Complete ad spots and product explainers in one pass
Video Output Price	$17.50 / 1M tokens	$12.00 / 1M tokens	-$5.50 / 1M tokens (-31.4%)	Substantial reduction in production rendering pipelines
Editing Workflow	Single-turn re-generation	Multi-turn conversational edits	Stateful persistence	Iterative parameter refinement without restarting
Audio Synchronization	Single voice baseline	Multi-speaker conditioning	Disentangled speech	Synchronized lip movements and comedy timing
Context Window	131,072 tokens	131,072 tokens	0 delta	Standardized multimodal developer context

Token metering economics across resolution tiers

Google structures video generation billing entirely through output token consumption rather than flat platform surcharges. Under this metering model, video rendering is billed according to output token generation velocity, which scales with resolution. At the published standard rate of $12.00 per million output tokens, generation expenses scale predictably:

Resolution token consumption rates and clip expenses under $12.00 per million output tokens

Resolution Tier	Token Rate (tokens/s)	Cost per Second (derived)	10-Second Clip Cost (derived)	15-Second Clip Cost (derived)
360p Draft	1,931 tokens/s	$0.023172 / s	$0.23172	$0.34758
720p Standard Native	5,792 tokens/s	$0.069504 / s	$0.69504	$1.04256
1080p Upsampled	8,688 tokens/s	$0.104256 / s	$1.04256	$1.56384
4K Master Upsampled	17,376 tokens/s	$0.208512 / s	$2.08512	$3.12768

Zoom

When evaluated against legacy Omni 1.1 Flash billing, developer expenses drop significantly. Under Omni 1.1 Flash, generating a standard 10-second 720p video consumed 57,920 output tokens at $17.50 per million, costing $1.0136 (derived). Under Omni 1.2 Flash, that same 10-second render costs $0.69504, saving $0.31856 per clip (derived), representing a 31.4 percent cost reduction (derived). A complete 15-second 720p native master costs $1.04256 (derived), allowing high-throughput marketing pipelines to generate broadcast-ready assets at approximately one dollar per sequence.

Multi-turn conversational video workflow and conditioning

The introduction of multi-turn conversational video editing restructures the cost profile of generative video pipelines. Instead of rendering complete sequences at full resolution during creative exploration, developers can execute a tiered pipeline:

Zoom

In Turn 1, developers generate an initial draft at 360p resolution for approximately $0.023 per second (derived). The model accepts multiple multimodal conditioning inputs simultaneously: front-and-back product photography, character reference sheets, short motion clips up to 3 seconds, and vocal audio samples. In Turn 2, conversational instructions modify environment parameters, such as changing a studio setting to an outdoor landscape, while preserving garment drapery, material reflections, and packaging typography. In Turn 3, multi-speaker dialogue tracks route distinct vocal inputs to individual animated characters with synchronized lip-sync. Only when the creative sequence is approved does Turn 4 trigger an upsampling pass to 1080p ($1.56 per 15s clip, derived) or 4K ($3.13 per 15s clip, derived).

Deprecation cutover, watermarking, and deployment limits

Google instituted an accelerated deprecation schedule for earlier model versions. Effective 3 December 2026, exactly 28 days following release (derived), all API requests addressed to gemini-omni-1.1-flash will automatically re-route to gemini-omni-1.2-flash and bill at the new $12.00 rate. This forced transition requires enterprise developers to audit existing prompt templates to ensure reference conditioning behaviors remain stable across the model upgrade.

To address content authenticity and provenance tracking, all visual frames and synthetic audio tracks synthesized by Gemini Omni 1.2 Flash incorporate imperceptible SynthID digital watermarking. SynthID watermarks persist through compression, format transcoding, and basic geometric alterations, enabling downstream verification platforms to authenticate AI-generated assets.

Stated capabilities versus acknowledged technical limitations in Omni 1.2 Flash documentation

Domain	Stated Capability	Acknowledged Limitation	Engineering Guidance
Camera Dynamics	Continuous FPV drone flights, orbital sweeps without jump cuts	Fast sudden scale changes can cause minor scale wobble	Constrain acceleration vectors in prompt instructions
Physical Simulation	Smoothed fluid dynamics and drapery motion	High-velocity sports action and rapid physical impacts remain challenging	Use multi-turn motion transfer for complex acrobatics
Dialogue & Audio	Disentangled multi-speaker routing with expressive timing	Rapid script transitions can cause pacing compression	Separate speaker turns with explicit timing markers
Deployment Tiers	Paid tier API, Google Cloud Vertex AI, Google Vids, Google Ads	Free tier video generation is completely unavailable	Budget token allocations using 360p draft estimates

While the model establishes transparent token metering and functional multi-turn editing, the release maintains strict commercial boundaries. Gemini Omni 1.2 Flash remains a closed proprietary cloud API. Google disclosed neither training dataset compositions, copyrighted media proportions, nor total compute budgets. Furthermore, all benchmark demonstrations published on launch day represent vendor-curated evaluations rather than independent third-party benchmark audits.