Gemini Omni Flash 1.2, which Google writes as Gemini Omni 1.2 Flash, arrived on 5 November 2026, ten weeks after 1.1. It is the same model shape: video with synchronised native audio, generated from text, stills, or short clips, then refined through follow-up prompts. The release is about ceilings rather than a new trick. A generation reaches fifteen seconds, an edit reaches the whole clip, and a second of 720p costs about 30% less. This page puts the two side by side.
Gemini Omni 1.2 Flash vs 1.1: the four changes that matter
Fifteen seconds instead of ten. A single generation now runs 3 to 15 whole seconds with ten as the default, where 1.1 capped at ten. Ten seconds is a fragment that usually needs extending before it reads as a scene; fifteen holds an approach, a turn, and an arrival, so more shots land in one pass.
Editing covers the whole clip. In 1.2 a follow-up applies across all fifteen seconds and can change the aspect ratio of an approved take, so the vertical cut of a landscape master is an edit rather than a second generation.
References are organised around people. 1.2 takes up to six reference images per request, three of them for each of up to two characters, where 1.1 advertised ten images. Reference video is now tied to output length: an 11 to 15 second clip takes one reference video of up to three seconds, while a 3 to 10 second clip takes up to three.
The price dropped. Google lists 720p at about $0.07 a second against roughly $0.10 for 1.1, with a 360p draft at about $0.023. A fifteen-second 720p clip lands near $1.04.
Specs side by side
| Omni Flash 1.2 | Omni Flash 1.1 | |
|---|---|---|
| Announced | 5 November 2026 | 27 August 2026 |
| Model id | gemini-omni-1.2-flash | gemini-omni-1.1-flash |
| Clip per generation | 3 to 15s, 10s default | 3 to 10s |
| Extension | 3 to 10s a turn, to 30s total | 10s a turn, to 40s total |
| Editing reach | The whole clip, including its aspect ratio | The same take |
| Reference images | 6 per request; 3 each for up to 2 characters | Up to 10 |
| Reference videos | 1 clip of 3s at 11 to 15s output; 3 clips below that | Up to 3 clips of 3s |
| Resolution | 360p, 720p native; 1080p, 4K upscaled | 360p, 720p native; 1080p, 4K upscaled |
| Aspect ratios | 16:9, 9:16 | 16:9, 9:16 |
| 720p price a second | About $0.07 | About $0.10 |
| 360p price a second | About $0.023 | Not published per resolution |
| Audio | Native, clearer lip sync, multi-speaker conditioning | Native, dialogue and music in one pass |
| Audio input | Unsupported in the current API | Unsupported |
| Watermark | SynthID | SynthID |
| On Morphic | Not yet | Available in the video model picker |
Resolution and aspect ratio are unchanged, which is worth saying plainly: if 720p at 16:9 was enough before, nothing about the output format forces a move.
What stays the same
Both models generate video with sound in one pass, take stills and short clips as input, interpolate between an opening and a closing frame, and keep refining a take through plain-language follow-ups. Both mark every output with SynthID. Both are English-first for dialogue and on-screen text, with other languages working unevenly. Neither accepts an audio upload or edits a voice through the API.
The weak spots carry over too. Google names fast sports action, scripts that change direction quickly, and complex physics as still in development in 1.2, and crowded scenes remain the place where tracked subjects drift in both. Editing or extending footage you uploaded yourself is unavailable in the EEA, Switzerland, and the UK.
One behaviour did change that will break old prompts: the angle-bracket tags that used to bind an image to a role are deprecated, and Google now asks for each file's job in natural language instead.
Which one should you use
If you are writing new work, 1.2, and the decision is barely a decision: it is longer, cheaper per second, and better at holding a face or a label through a camera move, with the same output formats. The reasons to care about 1.1 are historical rather than practical, since Google has said requests to 1.1 route to 1.2 from 3 December 2026.
The real constraint is where you are generating. 1.2 is in the Gemini API and rolling out through Google's own apps; it is not in the Morphic catalogue yet. 1.1 is, so the family's prompt shape and native audio are available on Morphic today, next to Veo 3.1, Seedance 2.5, and Kling, and a brief written for 1.2 mostly transfers: the structure, the audio direction, and the one-change-per-turn editing habit all carry. What does not transfer is the fifteen-second length and the reference arrangement.
Using Gemini Omni on Morphic today
Gemini Omni Flash 1.1 sits in the Morphic video model picker. Write the shot as a brief that includes the audio, generate, and the take lands on the Canvas, where the same brief can run through another model for a side-by-side read. Keep refining in follow-ups, one focused change at a time.
For the prompt mechanics, read the Gemini Omni Flash 1.2 guide. For the full specification, see the Gemini Omni Flash 1.2 model page or the Gemini Omni Flash 1.1 model page.
FAQs
On everything Google has published, yes: a generation runs to fifteen seconds instead of ten, edits apply across the whole clip, consistency through camera moves is described as steadier, and a second of 720p costs about 30% less. Output formats are identical, so there is no resolution or aspect-ratio trade-off to weigh.
About 30% a second. Google lists 720p at roughly $0.07 against about $0.10 for 1.1, which puts a ten-second clip near $0.70 and a fifteen-second one near $1.04. Drafting at 360p costs about $0.023 a second, and 1080p and 4K are billed higher as upscales.
Google has said that requests to the 1.1 model route automatically to 1.2 from 3 December 2026 and are billed at 1.2 rates. That is a redirect rather than a shutdown announcement, but it means new work has no reason to target 1.1 on the API.
Fewer images, but arranged more usefully. 1.2 documents six reference images per request with three each for up to two characters, where 1.1 advertised ten. Reference video is now tied to the length you are generating: one clip of up to three seconds for an 11 to 15 second output, up to three clips below that.
The prose does. The tags do not: the angle-bracket tokens that bound an image to a role are deprecated, and Google asks for each file's role in natural language instead, such as using the first image as the opening frame. Duration is the other thing to revisit, since a prompt written for ten seconds will under-fill fifteen.
Not yet. Gemini Omni Flash 1.1 generates on Morphic today with native audio, alongside Veo 3.1, Seedance 2.5, Kling and the rest of the catalogue, and Morphic has a free plan to try it on.