Gemini Omni 1.2 Flash Is Generally Available at $0.07 per Second of 720p Video. A Maximum 15-Second Clip Costs $1.05, and That Is Your Margin Problem.
Google made Gemini Omni 1.2 Flash generally available on Google Cloud on November 6, with clips up to 15 seconds and a price of $0.07 per second of 720p video. Multiply it out and a maximum-length clip costs $1.05 before a single retry. If you're thinking about a video feature in your product, that number decides the pricing page.
What shipped
The Google Cloud post says the model generates and edits video through the API and the console. Clips went from 10 to 15 seconds. You can draft at 360p or 720p and upscale to 1080p or 4K. Editing is conversational: swap a product, change the lighting or the camera angle. Speech and multi-speaker dialogue improved, and the model accepts image and video references. Outputs carry a SynthID watermark, Provisioned Throughput is offered, and audio references are gated behind allowlisting. Gemini for business and Google Vids integration is listed as coming soon.
The price is 30% below Gemini Omni 1.1 Flash, which by my arithmetic puts the old price near $0.10 per second. The post says similar savings apply at other resolutions, and points to the Model Garden pricing page for the full table. I haven't read that table, so I don't have 360p, 1080p or 4K prices.
The math that matters is per accepted clip
Per second is the vendor's unit. Yours is per clip a customer keeps. A 15-second 720p clip is $1.05. If a user keeps one clip in three, which is a guess and not a measured rate, the cost of one kept clip is about $3.15. Put a user on a $19 monthly plan who generates ten maximum-length clips, and you've spent $10.50 on generation alone, before upscales, storage, payment fees and support. At a one-in-three keep rate, those ten clips are three keepers and a margin that is already thin.
The post doesn't say how conversational edits are billed. If every edit regenerates a clip at the per-second rate, editing is where the bill hides, so I'd test that on a small budget before promising "unlimited tweaks" to anyone.
What I'd build around it
Sell credits, not unlimited. A credit is a 5-second 720p draft, a final render costs more credits, and the customer sees the cost before they press the button. Let people draft cheap and upscale only what they keep. Cache and reuse anything the same customer asks for twice. And set a hard daily spend cap per account in your own code, because a retry loop in a client app can burn money faster than any invoice alert will tell you.
The watermark matters too. SynthID is built in, so if a customer expects unmarked output for commercial use, find that out in week one, not after launch.
Where I could be wrong
I haven't compared this price with other video providers, so treat $0.07 as a number to benchmark, not a verdict. Prices in this category have moved a lot in a short time; I wrote earlier that LLM prices alone changed eight times in nine weeks, and I'd expect video to do the same. A 30% cut may be followed by another before your feature ships. If your whole margin depends on today's price, that is the real warning.
What I'd actually do
Prototype with 5-second drafts at the lowest resolution that looks acceptable, log every generation with its cost and whether the user kept it, and look at two weeks of data before you set a price. If the keep rate is under 30%, fix the prompt flow before you fix the pricing. If it's over 60%, you can afford to be generous and still sleep. Either way, the number you want on a dashboard is dollars per kept clip.
Author
Lukas
@lukcombinator