Gemini Omni 1.2 Flash launches with 15-second video generation and 30% price cut
Google DeepMind releases Gemini Omni 1.2 Flash with longer video clips, improved consistency, and lower pricing across the Gemini API.
Last verified:
Gemini Omni 1.2 Flash: Lower cost, longer format, tighter consistency
According to Google AI Blog, Google DeepMind released Gemini Omni 1.2 Flash on November 5, implementing a 30% price reduction to $0.07 per second for 720p video generation and extending native video creation from 10 to 15 seconds. The model is now available across the Gemini API and Gemini for business, positioning it as a more accessible option for video-generation workflows at enterprise and developer scale.
The pricing reduction is paired with architectural changes that lower computational drag across the generation pipeline. Rather than forcing users to re-generate entire clips when adjusting specific details, Gemini Omni 1.2 Flash supports multi-turn conversational editing—developers and creators can draft at 360p or 720p, iteratively refine via follow-up prompts, and then upscale the final output to 1080p or 4K. This tiered approach reduces the per-output cost relative to single-pass full-resolution generation.
Extended duration and granular control
The extension from 10-second to 15-second native video clips removes a structural barrier for use cases requiring narrative arc: product explainers, social-media ad spots, and short-form branded content. Importantly, Gemini Omni 1.2 Flash allows duration specification in 1-second increments between 3 and 15 seconds, avoiding the binary “fit-or-bust” problem where users previously had to pad or truncate scripts to match fixed clip lengths.
Reference adherence and visual consistency
The most significant technical improvement is stronger reference following. According to Google DeepMind, the model can now ingest character images, product photography, motion clips (up to 3 seconds), and audio cues, then maintain face identity, garment details, packaging typography, and brand colors across camera transitions, depth changes, and full-sequence generation. This addresses a long-standing friction point in commercial video production—the need to composite multiple model outputs or manually fix visual drift frame-by-frame.
The consistency gains extend to scene and object stability: characters, props, and environments remain visually coherent from first to final frame, with reduced morphing during camera moves or shot transitions. Google DeepMind demonstrates this with examples spanning virtual try-on sequences, 360-degree product spins, and animated poster generation, all maintaining reference-image fidelity across the full clip.
Why This Matters
The 30% price reduction directly impacts unit economics for teams building video-generation workflows at scale—marketing agencies, e-commerce platforms, and in-house creative ops now face lower per-second costs, making batch video production feasible at volumes previously reserved for specialized studios. The multi-turn editing loop further reshapes creative iteration: teams can experiment at low resolution, refine via conversation rather than new prompts, and finalize at high quality, compressing the design-to-delivery cycle.
For product-centric use cases (catalog videos, virtual try-ons, brand consistency enforcement), the reference-adherence improvements reduce the manual stitching and rotoscoping overhead that currently requires human intervention. If the consistency holds under production load, this moves Gemini Omni 1.2 Flash closer to a viable replacement for templated and semi-manual video pipelines in commerce and marketing.
Frequently Asked Questions
What are the key pricing and capability changes in Gemini Omni 1.2 Flash?
Pricing dropped 30% to $0.07/second for 720p video (from Omni 1.1), maximum clip length extended from 10 to 15 seconds with 1-second granularity, and consistency improvements for character and product references.
How does the multi-turn editing workflow reduce costs?
Users can draft at lower resolution (360p), edit specific details iteratively rather than re-generating entire clips, then upscale final versions to 1080p or 4K, avoiding wasteful full re-renders.
What reference materials can guide video generation?
The model accepts character, product, or style images; motion clips up to 3 seconds long; and audio cues to maintain consistency for faces, garments, packaging, typography, and colors throughout the generated video.