Gemini Omni 1.2 Flash API: model ID, pricing, and the free routes

Gemini Omni 1.2 Flash is Google DeepMind's video generation and editing model, announced November 5, 2026 and generally available the same day across the Gemini API and Gemini for business [official]. It is the first Omni version to reach general availability, backed by Google Cloud's enterprise infrastructure, with Provisioned Throughput for high-volume work [official].

One model does four jobs: text-to-video, image-to-video, conversational editing of a clip you already made, and standalone upscaling to 1080p or 4K [official]. Every clip carries native audio — dialogue, music and sound effects generated in the same pass [official] — and an imperceptible SynthID watermark [official]. You guide it with character, product or style images, short motion clips up to three seconds, and audio cues, and it holds those references from the first frame to the last [official].

The three numbers that matter: clips run 3 to 15 seconds, in one-second steps, up from 10 seconds on Omni 1.1 [official]; the price is $0.07 per second at 720p, 30% below Omni 1.1 [official]; and the model ID is gemini-omni-1.2-flash [official]. Google says it replaces Veo in the Gemini app, and the Veo 3.1 preview models on the Gemini API were shut down on October 22, 2026, with Omni Flash as the named replacement [reported].

Step 1 — a paid-tier API key. Create a key in Google AI Studio for a project with billing turned on [reported]. There is no free tier for this model: Google's pricing page lists the free tier as "Not available" for both input and output [official]. A key from a free-tier-only project will not run Omni.

Step 2 — call the Interactions API, not generateContent. Omni does not run on the generateContent endpoint the text and image models use [reported]. It runs on the Interactions API — POST to the interactions endpoint with the model ID gemini-omni-1.2-flash [reported]. Every generation returns an interaction ID, and a follow-up edit references the previous interaction instead of re-sending the whole brief [reported].

Step 3 — set duration, resolution, aspect. Duration is any whole number of seconds from 3 to 15 [official]. Resolutions are 360p, 720p, 1080p and 4K, with 1080p and 4K reached by upsampling [official]; aspect ratio is 16:9 or 9:16 [reported]. Reference rules from the developer guide: up to six reference images across durations, and for video references, an 11–15 second video accepts one reference video up to three seconds long, while 3–10 second videos accept up to three [reported].

Step 4 — edit in conversation, extend, upscale. Refine the same clip turn by turn — swap a product or background, change the lighting or weather, switch the aspect ratio — instead of starting over [official]. Extend a clip by 3 to 10 seconds per turn, up to 30 seconds total [reported]. Upscaling is its own step: any clip up to 15 seconds long can go to 1080p or 4K without being made again [reported].

This page was written from Google's announcement, pricing page and developer guide, plus independent write-ups that read them — no live API generation was run for it, so there is no hands-on quality verdict here. The code shape above is the documented shape, not a tested snippet.

Route 1 — Google Vids (the only free route with 15-second clips). Google Vids makes 15-second clips at no additional cost with a Google account or Workspace subscription [official]. No API key, no billing setup. The exact monthly allowance in Vids could not be verified — independent write-ups disagree on the number — so check the Vids page before planning around a quota [unknown].

Route 2 — Google Flow (free daily credits, shorter clips). Flow gives free daily credits with or without a subscription, and lists Omni Flash 1.2 generation and editing up to 10 seconds for all users [reported]. Ten seconds, not fifteen, is the free-user ceiling here [reported]. Flow is the fastest way to feel the conversational edit loop without paying — and the paid plans extend the clip length and the credit pool.

Route 3 — the Gemini app (paid plan only). The model is rolling out in the Gemini app to Google AI Ultra subscribers [official]. That is a subscription route, not a free one — it belongs here only so you do not mistake it for a free tier.

The split to remember: free means Flow or Vids, paid means the API. There is no free API tier to unlock with a different key or a different endpoint.

Google bills Omni by the token, not by the second — but it prints the per-second conversion on the same page. Video output costs $12.00 per million tokens, at 5,792 tokens per second of 720p video, which works out to approximately $0.07 per second [official]. A 10-second 720p clip is about $0.70 of video output, a 15-second clip about $1.04 [official].

The other resolutions from the same official page: $0.023 per second at 360p (1,931 tokens per second), $0.104 at 1080p with upsampling (8,688 tokens per second), and $0.209 at 4K with upsampling (17,376 tokens per second) [official]. A 15-second clip is therefore about $0.35 at 360p, about $1.56 at 1080p and about $3.13 at 4K, video output only [official]. Input costs $1.50 per million tokens for text, image and video, and text output is $9.00 per million [official]. Audio carries no separate charge [reported].

Against Omni 1.1: video output was $17.50 per million tokens, about $0.10 per second at 720p [official] — the 30% cut in the announcement [official]. And the migration date: effective December 3, 2026, all API requests to gemini-omni-1.1-flash automatically route to gemini-omni-1.2-flash and are billed at the lower 1.2 rates, with no code change [official].

This is an early bet, not a proven search term. The model shipped November 5, 2026. On November 8, Google autocomplete returned no suggestions for "gemini omni 1.2 flash api", "gemini omni 1.2 flash free" or "gemini omni 1.2 flash pricing" [reported] — the version-specific phrasing has no established demand yet. The family-level phrase "gemini omni flash api" does autocomplete, with pricing, key and documentation suffixes [reported], so the API intent is real and the 1.2 wording is the early part.

The quality claims are Google's, with no benchmarks behind them. Better reference adherence, steadier camera motion, more natural dialogue and lip-sync — all from the launch post, with no published benchmark numbers [official claims; no benchmark reported]. One independent tester ran 20 API tests and called it the most instruction-exact video model they had tested, strong on explainers and typography, weaker on photoreal people and cinematic brand films [reported] — one tester's verdict, not a leaderboard.

Google names its own weak spots. The launch post says development is ongoing on fast sports action, rapid script changes and complex physics [official]. Each edit turn also re-synthesizes the clip rather than editing a timeline, so "conversational editing" still means a new video each turn.

Two limits to check before you build. Editing or extending uploaded videos is unavailable in the European Economic Area, Switzerland and the UK, per the developer guide [reported] — a regional restriction that does not apply to fresh generations. And the free-route quotas are the least verified numbers on this page: the Vids monthly allowance is unconfirmed [unknown], and Flow's free-credit figures come from secondary write-ups of Google's credits page, not from a page this guide opened directly [reported].