Gemini Omni 1.2 Flash: 15-Second Clips at 7 Cents a Second

On November 5, 2026, Google made Gemini Omni 1.2 Flash generally available on the Gemini API and Google Cloud. Same day announcement and GA. It is the first Omni model generally available on Google Cloud; on the Gemini API, Omni 1.1 Flash had already been GA since August 27, 2026.

Omni is Google's video generation and conversational editing model with native audio. You prompt with text, images and video; it returns video with speech, music and sound effects, plus text. It runs on the Interactions API only. Google positions it for the everyday videos businesses make in volume: product clips, virtual try-ons and ads. See also our coverage of Google's image model Nano Banana 2.1.

The release is a price and production update, not a new kind of model. Video output drops from $17.50 to $12.00 per 1M tokens, so 720p lands at about $0.07 per second ($4.17 per minute of output). Clips grow from 10 to 15 seconds. From Dec 3, 2026 every 1.1 request auto-routes to 1.2 at the lower rate. The bill-moving choice is resolution: 4K costs 9x 360p per second.

At a Glance

What Changed Since Omni 1.1

Lineage is short. gemini-omni-flash-preview opened public preview on June 30, 2026 (3 to 10 s, 720p only) and is now shut down. gemini-omni-1.1-flash went GA on the Gemini API on August 27, 2026 with extension, first and last frame interpolation, and 360p to 4K with upscaling. On Google Cloud, 1.1 is still a Preview (gemini-omni-1.1-flash-preview). 1.2 is the production cut.

Longer clips

Native generation and editing move from 10 s to 15 seconds, settable from 3 to 15 s in 1 s steps. Photoroom, quoted by Google Cloud, called 15 s videos "a top user request." QuickFrame AI says 15 s at 1080p lets users create a complete CTV ad in a single generation. First and last frame interpolation still tops out at 10 s; longer with two frames returns an invalid argument error. Extension remains 3 to 10 s per turn (default 10), up to 30 s total, end of clip only.

A 30% price cut

Google says 720p is about $0.07 per second, 30% lessthan Omni 1.1 Flash, "with similar savings at other resolutions." The token rate behind that is $12.00 per 1M video output tokens, down from $17.50. Input and text output prices are unchanged. Full breakdown in Pricing.

Reference adherence and consistency

Google says reference adherence is better: up to 6 reference images (up to 3 per character, up to 2 characters), plus short motion clips up to 3 s. Faces, garments, packaging typography and brand colors stay consistent across camera angles, per Google. Google also claims less drift or morphing from first to last frame, including across interpolation and extension. It shows side-by-side demos; it published no numbers.

Editing in conversation across full 15 s clips

Multi-turn editing runs through the Interactions API with previous_interaction_id, across full 15 s clips, without re-uploading. Google demos include swapping a product or background, changing lighting or weather, and switching aspect ratio.

Speech and camera

Google says speech is more natural: more fluid dialogue, clearer lip-sync, and distinct multi-speaker conditioning. On camera, it claims complex continuous moves (FPV flythroughs, zoom-outs, orbits) with "significantly fewer unwanted jump cuts." No numeric quality deltas.

Production plumbing

This is the first Omni model generally available on Google Cloud, with Provisioned Throughput there (the Gemini API guide says Provisioned Throughput is not supported). Standalone upscaling is new on Google Cloud; 1.1 lacked it there. Paid-tier Gemini API rate limits are doubled. Photoroom's quote on the Google Cloud post mentions general availability and SLAs.

What did not change: same input and text output prices ($1.50 / $9.00 per 1M), same resolutions and aspect ratios, same Interactions API, same 131,072 input limit, still no Batch and no free tier.

Pricing: What a Second, a Clip and a Minute Cost

Official rates from the Gemini API pricing page and Google Cloud generative AI pricing. Token rates are identical across both. Video is billed as output tokens per second of video with audio.

Per-minute figures are 60 seconds of output, computed from Google's per-token rates (a single generation tops out at 15 s). Omni 1.1 values other than 720p are computed from Google Cloud's published token-per-second counts at 1.1's $17.50 rate. $12.00 vs $17.50 per 1M is about 31% lower; Google rounds to 30%. 4K is 9x 360p and 3x 720p per second.

Worked examples

• A 15 s 720p social ad: 15 × 5,792 = 86,880 tokens = $1.04(Google says "~$1.04").
• One minute of 1080p, e.g. four 15 s clips: $6.26 vs $9.12 on 1.1 (computed).
• One minute at 4K: $12.51, which is 9x the $1.39 for a minute at 360p.
• Iteration on Google Cloud rates: four 15 s 360p drafts plus a 1080p upscale of the winner = $2.44, vs $6.24 for four direct 1080p takes.
• Inputs are small: about $0.0017 per reference image (1,120 tokens at $1.50/M), so six references cost about a cent. Sending a 15 s clip to edit costs up to about $0.13 in input tokens (up to 5,792 tokens per input video second).

Thinking tokens bill at the $9.00 text rate. Google does not say how many a video request uses.

Draft at 360p or 720p, then upscale keepers. On Google Cloud rates, upscaling never costs more than rendering the target resolution directly. The Gemini API docs describe 1080p and 4K as upscaled via resolution but do not list a separate upscale rate; the draft-then-upscale math above uses Google Cloud rates.

Other video models we have covered, without comparing prices here: WAN 3.0 and Seedance 2.5.

When to Use It, and Migrating from 1.1

Use it if you produce short video in volume: product spins, virtual try-on, merchandise clips, social and CTV ads, explainers. Anyone already on 1.1 should plan for Dec 3, 2026, when routing is forced.

• You do not have to change code. From Dec 3, 2026, gemini-omni-1.1-flash routes to gemini-omni-1.2-flash and bills at 1.2 rates.
• Switch the id now to get 15 s and the lower price about four weeks early.
• Re-test prompts that rely on exact 1.1 behavior. Output will change after the routing date.
• Set duration explicitly if you want more than 10 s.
• Keep two-frame interpolation at 10 s or less. Longer returns an error.
• Budget for 11 to 15 s clips taking only 1 reference video (3 for clips up to 10 s).
• Draft at 360p or 720p and upscale winners.
• On Google Cloud, 1.1 was a Preview, so 1.2 is the version to put behind an SLA.

Switch now. There is no price or capability reason to wait for the auto-route.

Caveats

• Google published no benchmark scores or numeric quality comparisons for Omni 1.2 Flash; its quality claims are its own and have not been independently verified.
• No generation speed or latency numbers. Google gave none.
• Google says fast sports action, rapid script changes and complex physics still need work.
• Audio references are not open to everyone: the Gemini API does not accept them, and Google Cloud gates them behind an allowlist. Voice editing is not supported.
• English is the only fully supported language; others are not evaluated.
• Editing or extending uploaded videos is unavailable in the EEA, Switzerland and the UK (model-generated videos can be edited and extended). Uploading or editing images of minors is unsupported there. Images of certain recognizable people are unsupported.
• No Batch. No free tier. No system instructions, temperature or negative-prompt parameter (put negatives in the prompt).
• Gemini API and Google Cloud documents differ on output token limits (65,536 vs 260,640) and Provisioned Throughput (unsupported vs supported).

FAQ

Outlook

Things to watch, without date predictions: whether audio references open up (today the Gemini API does not accept them, and Google Cloud gates them behind an allowlist); Gemini for business availability (Google says it is coming); Batch or a cheaper tier for bulk jobs; longer native clips past 15 s; and independent quality measurements. For now, the release you can act on is the one already shipping: lower video tokens, 15 s clips, and Google Cloud GA.

Sources: Google DeepMind announcement, Google Cloud GA post, Gemini API model page, Gemini API pricing, Omni guide, rate limits, Google Cloud model page, Google Cloud pricing.