Think of how a director works: shoot a take, watch it, say "again, but slower, and keep the red coat". Gemini Omni 1.2 Flash brings that loop to the API. It generates video with native audio from text, images or existing video, then lets you refine the result turn by turn through the Interactions API instead of starting each prompt from scratch.
It reached general availability on 5 November 2026 and is now Google's named replacement for its earlier preview video models.
Specification
| Field |
Value |
| API ID |
gemini-omni-1.2-flash (stable) |
| Released |
5 November 2026 |
| Input |
Text, image, video |
| Output |
Video with native audio, plus text |
| Input token limit |
131,072 |
| Output token limit |
65,536 |
| Clip length |
3 to 15 seconds |
| Extension |
3 to 10 seconds per turn, up to 30 seconds total |
| Resolution |
360p, 720p, 1080p, 4K |
| Frame rate |
24 fps |
| Reference images |
Up to 6 |
Video input can run up to 15 seconds for editing, with up to 10 seconds of lookback for extension. First- and last-frame interpolation lets you pin where a shot begins and ends.
Consistency across shots
The six-reference-image limit is the feature that matters most for anyone making more than one clip. Keeping a character's face, a product's shape or a mascot's colours stable across shots has been the weak spot of generated video. Google says 1.2 improves character and subject consistency across those references. Test it with your own assets, because consistency claims are easy to make and hard to measure.
What a clip costs
Pricing is per token. Input of any type is $1.50 per million. Video output is $12 per million, 30% lower than 1.1. Text output is $9 per million. Video is counted at 5,792 tokens per second of 720p footage, which Google converts into rough per-second prices:
- 360p: about $0.023 a second
- 720p: about $0.07 a second, so roughly $0.70 for ten seconds and $1.04 for fifteen
- 1080p with upsampling: about $0.104 a second
- 4K with upsampling: about $0.209 a second
There is no free tier. Some gateway listings show text output at $9 and others at $12; Google itself prices text at $9 and video at $12, so a $12 text rate probably conflates the two. Check which rate your provider applies. All figures are indicative.
Paid-tier rate limits doubled compared with 1.1.
Migration deadlines
Three preview video models, Veo 3.1 Preview, Veo 3.1 Fast Preview and Veo 3.1 Lite Preview, were deprecated with a 22 October 2026 shutdown and point here. So does Gemini Omni Flash Preview, shut down on the same date. Anyone on gemini-omni-1.1-flash has less to do: requests to that ID route to 1.2 automatically from 3 December 2026. If your pipeline is sensitive to output changes, pin to the 1.2 ID now and test before the switch happens for you.
A storyboard test
Write a five-shot sequence with one recurring character. Supply three reference images. Generate each shot at 720p, then extend the best one twice. Score identity drift between shots, audio sync, and how many turns each shot needed. Multiply turns by the per-second price and you have a realistic cost per finished second, which is usually two or three times the headline figure.
Other generators are listed under text-to-video and image-to-video.