Gemini Omni 1.2 extends clips to 15 seconds and cuts video-output prices

Google released Gemini Omni 1.2 Flash on November 5, extending native video generation from the previous version’s ten-second ceiling to fifteen seconds. It also announced a 30% reduction in the 720p video-output rate, to roughly $0.07 a second. Video makers can generate longer scenes and request follow-up edits. Google says follow-up prompts can change backgrounds or lighting.

Video-API provider Segmind published a first-hand test involving seventeen generations and a comparison clip. Its author reported charges of $1.069 for ten seconds at 720p on 1.1 and $0.741 on 1.2. Reference-only mode and first-and-last-frame input hit a ten-second limit in that provider’s tests. Those results describe its service, rather than universal Google API behavior.

The author also noted soft background faces in a wide night-market scene and declined to declare a quality winner from one matched pair.

The fifteen-second maximum depends on the input mode. Gemini API documentation limits interpolation between a supplied first and last image to three to ten seconds. Generations lasting eleven to fifteen seconds accept one reference video of up to three seconds; up to six reference images are supported.

Editing or extending uploaded videos is unavailable in the EEA, Switzerland and the UK. The guide says follow-up editing and extension of model-generated videos remain supported there.

Video-output prices exclude input charges

The Gemini API pricing page lists paid-tier access, with no free API tier for this model. Standard rates per million tokens are $1.50 for text, image and video input, $9 for text output and $12 for video output. At 5,792 video tokens a second, fifteen seconds of 720p output costs approximately $1.04.

Other listed video-output rates are approximately $0.023 a second at 360p, $0.104 at 1080p and $0.209 at 4K. The latter two are upsampled outputs.

Google Cloud’s model card gives November 5 as the GA release date. It supports text and video output, including video with generated speech, music and sound effects. Standalone audio input is listed as unsupported. Generating a soundtrack and supplying an audio file as a reference are separate capabilities.

The Cloud offering supports the Interactions API and Provisioned Throughput, with global model availability. The developer Gemini API is a separate offering. The Cloud card also lists C2PA Content Credentials for communicating content provenance.

App allowances are separate again. Vids’ current account table lists sixty AI-video seconds a month for personal accounts and 500 for Google AI Pro. Work and school administrators control Vids and generative-AI access; unavailable features do not appear on the start screen. Basic access to the editor does not remove generation limits. Business Starter has 200 seconds monthly; Standard and Plus have 500. These are current account allowances, not a newly announced across-the-board increase. Google AI Ultra 5x lists 2,500 seconds monthly, and 20x lists 10,000.

Generated videos include Google’s invisible SynthID watermark, embedded during creation and designed to tolerate changes including cropping, filters and lossy compression. SynthID technology detects the mark. Google applies watermarking to images, video, audio and text. Existing Gemini API requests for Omni 1.1 will automatically route to 1.2 after the four-week transition window, on December 3, at the lower rates.