Video Generation Model
also called Text-to-video model, Video model
A model that turns text, images or video into new moving footage with sound.
A model that generates video clips from a prompt — text, a reference image, or an existing clip — with motion, synchronized audio and dialogue, and that can edit the clip in later turns instead of starting over. The clip is the unit of output: a few seconds long, priced by the second, and increasingly the raw material for ads, explainers and product shots rather than a finished film.
Google’s Gemini Omni 1.2 Flash, generally available from November 5, 2026, shows where the category has settled: native 15-second clips (up from 10), duration settable in one-second steps from 3 to 15, multi-turn conversational editing across the whole clip, reference images and short motion clips held consistent from first frame to last, and native audio with lip-synced multi-speaker dialogue. The price fell 30% from the previous version, to $0.07 per second at 720p — about a dollar for a 15-second clip.
The workflow around the model matters as much as the model. You draft cheap at 360p, fix a detail with a follow-up edit instead of re-generating, and upscale the keeper to 1080p or 4K — the higher resolutions are upscaled, not generated natively. And the output carries an imperceptible SynthID watermark, so provenance is built into the clip rather than declared alongside it. Google published no benchmarks for the release, so the quality claims are the vendor’s, not a leaderboard’s.
The price structure is the adoption story. At $0.07 per second of 720p, a 15-second clip costs about $1.04 on the Gemini API — cheap enough that the cost of a video ad is now dominated by the prompting, review and iteration around it, not the generation. Google prices by output tokens ($12 per million) and converts to seconds at a fixed token rate per resolution, so 1080p and 4K cost more per second even though they are upscales of a 720p-or-lower generation.
General availability is the other half of the news. Omni 1.2 Flash is the first model in the Gemini Omni family to leave preview, backed by Google Cloud with Provisioned Throughput for production workloads — and the old version is already on a clock: Google routes 1.1 requests to 1.2 from December 3, 2026, and Flow retires the 1.1 and Veo 3.1 models on December 5. A video pipeline built on 1.1 is a migration, not a stable dependency.
The product shot that became a spot
Marketing wants a 15-second video ad from existing product photography, with three outfit changes and a consistent model.
Feed one model photo plus front-and-back shots of the three outfits and ask for a 15-second virtual try-on sequence: Omni 1.2 Flash holds the model’s identity and each garment’s cut and drape across 360-degree turns and match cuts. If the background is wrong, edit it in a follow-up turn instead of re-generating — the model edits the clip you have. Draft at 360p, upscale the keeper, and check the SynthID watermark survives your export pipeline before the spot ships.
Treating a generated clip as footage. It is a synthetic artifact with a watermark and a short native length; extending it or cutting it into a longer sequence is a second job, and fast action and complex physics are still weak spots.
- blogGemini Omni 1.2 Flash: 15-second video and lower pricing ↗
The GA announcement: native 15-second clips with synchronized audio and multi-turn conversational editing, 30% cheaper at $0.07 per second of 720p, with a SynthID watermark on every clip. Quality claims are Google’s own; no benchmarks were published.
Diffusion Model
An image generator that learns to reverse a process of adding noise.
Multimodal
A model that handles more than text — images, audio, video, screen pixels.
Model Provider
The company or platform you call to run inference.
Cost per Task
What one unit of real work actually costs.