Google Gemini Omni 1.2 Flash 正式可用，原生片段延长至 15 秒

Google announced Gemini Omni 1.2 Flash on November 5, 2026, making the video-generation and editing model generally available. The update raises native clip length to 15 seconds, adds multi-turn conversational editing and cuts 720p pricing by 30% to approximately $0.07 per second.

The model generates synchronized video and native audio from text, image and video inputs. The additional running time is intended to fit complete stories, product explainers and full social advertising spots into a single generation.

Longer Clips and Conversational Editing

The developer documentation supports whole-second durations from 3 to 15 seconds, compared with a 10-second maximum for Omni 1.1, and uses 10 seconds as the default. Output defaults to 720p and 16:9 landscape, with 9:16 portrait also supported. Teams can draft at 360p or 720p and upscale selected results to 1080p or 4K.

Conversational editing can replace a product or background, alter lighting or weather, or switch aspect ratio while leaving unmentioned elements intact. A previous-interaction identifier preserves context between API requests, so the prior video does not have to be uploaded again. Each request can also use as many as six reference images, including up to three images per character for up to two characters.

Video extension is separate from native generation: it appends 3 to 10 seconds per turn, using the final 10 seconds as context, up to 30 seconds total. It cannot prepend footage or insert material mid-clip, and first-to-last-frame interpolation remains limited to outputs of 3 to 10 seconds.

Pricing and Migration

On the paid Gemini API, pricing is $12 per million video-output tokens, metered at 5,792 tokens per second of 720p video. The listed costs are approximately $0.70 for a 10-second clip and $1.04 for a 15-second clip. Text, image and video input costs $1.50 per million tokens, and the model is not available on the free tier.

Gemini Omni 1.1 Flash remains available during a four-week transition. On December 3, 2026, API requests to 1.1 will automatically route to 1.2 without code changes and use the lower 1.2 rates. Existing integrations therefore have a defined migration date without an immediate rewrite.

Quality Claims and Production Limits

Google says the update better preserves faces, garments, packaging typography and brand colors across camera angles and transitions, with less visual drift or morphing, while delivering clearer lip-sync, more natural dialogue, distinct conditioning for multiple speakers and continuous camera movement with fewer unwanted jump cuts.

Google DeepMind’s side-by-side evaluations give Omni 1.2 a 62% visual-consistency win rate against 38% for Omni 1.1 in text-to-video testing. The reference-to-video comparison gives 1.2 a 63% multi-reference identity-consistency win rate against 37%.

There are also explicit technical limits. Development continues on fast sports action, rapid script changes and complex physics. The Gemini API fully supports English but has not evaluated other languages. Voice editing, uploaded audio inputs and audio references are unsupported in the current API. Generation time varies with duration, resolution and API load, and safety filters apply to prompts and generated video.

Availability and Production Use

Google Cloud’s general-availability announcement adds console and API access, enterprise infrastructure and security, and Provisioned Throughput, with audio references gated behind allowlisting. Photoroom chief executive Matt Rouif said the longer generation limit was a top user request and that general availability with service-level agreements would let the company rely on Gemini Omni Flash in production for sellers and enterprise customers. The model powers Photoroom’s Video Ad Maker and Product Animator.

The release links lower pricing to a changed production loop: targeted revisions can replace full regenerations, and higher-resolution rendering can be reserved for selected outputs. Its production value therefore rests on the reported consistency gains, since preserving characters, products and scenes across revisions is what makes iterative editing a practical alternative to starting over.