A longer video clip changes the edit; an automatic model switch can change the whole production pipeline.
Google announced Gemini Omni 1.2 Flash on November 5, adding native video generation up to 15 seconds and conversational editing. The model is generally available on the paid Gemini API tier. Google also published a December 3 change for applications still requesting Omni 1.1 Flash.
Match the duration to the operation
The API release notes distinguish fresh generation from other operations. First-and-last-frame interpolation supports 3 to 10 seconds. A video extension adds 3 to 10 seconds per turn, up to 30 seconds in total. A request that fits one of those paths may exceed another path’s limit.
Google describes stronger reference adherence and more consistent subjects across a clip. Those are company claims. We have not compared generated scenes or measured how often a character, product label, or camera movement stays correct.
A 15-second output costs about $1.04 at 720p
Google’s standard API price table lists video output at US$12 per million tokens. It states a conversion of 5,792 tokens per second for 720p. Multiplying those figures gives US$1.04256 for 15 seconds, before input tokens or other charges.
| Clip length | Calculated 720p video-output charge |
|---|---|
| 5 seconds | US$0.34752 |
| 10 seconds | US$0.69504 |
| 15 seconds | US$1.04256 |
Three 15-second attempts therefore carry about US$3.13 in video-output charges before any editing or input costs. For a production budget, record how many attempts it takes to get an accepted clip. A lower price per generation can still leave substantial review and repair work.
The old model name will change its destination
From December 3, Google says requests to gemini-omni-1.1-flash will route automatically to gemini-omni-1.2-flash at the new rates. Teams using a fixed model name should therefore record a separate migration date in their release plan.
Keep a small set of approved prompts and reference assets for comparison before that switch. Include an interpolation request and an extension if the application uses them. Higher-resolution output also deserves its own cost review: the documentation distinguishes upsampling from native detail.
Our Omni 1.1 guide retains the earlier release history and the migration context. Which matters more for your next video workflow: a longer first generation or fewer retries to keep the subject consistent?
Abdessalam Alaoui
I'm Abdessalam Alaoui, founder of Musthave.AI. I write practical guides to AI models, tools, generative media.
