Gemini Omni Flash Prices Video at 10 Cents a Second With No Free Tier. If Your Idea Was 'AI Video, But Cheap,' Run the Math First.
Gemini Omni Flash Prices Video at 10 Cents a Second With No Free Tier. If Your Idea Was 'AI Video, But Cheap,' Run the Math First.
Google made Gemini Omni Flash generally available at the end of June: conversational video generation and editing, in the Gemini API, AI Studio, Flow, and the Enterprise Agent Platform. The demos are genuinely good. The pricing is where solo builders need to stop and do arithmetic: roughly $0.10 per second of 720p video output, no free tier, no batch discount. Input (text, image, video, or audio references) is a flat $1.50 per million tokens. The current clip cap is 10 seconds.
Do that multiplication once. A single 10-second 720p clip costs a little over a dollar in output alone, before input. That's the same neighborhood as Veo 3.1 Fast, so this isn't Google being greedy: it's the actual floor for frontier video generation right now. And if your product thesis was "AI video, but cheaper than the incumbents," that floor is the thing you're about to run headfirst into.
The number that kills consumer video wrappers
Let's price a plausible solo-operator product: a consumer app that turns a text prompt into a short social clip. Say the average user generates 20 clips a month, each 8 seconds, and half of them get regenerated because the first take wasn't right. That's 30 generations at 8 seconds (240 seconds of output), roughly $24 in raw generation cost per active user per month, before you've paid for input tokens, storage, egress, or yourself.
Now try to charge for it. A $9.99/month consumer subscription doesn't survive a $24 cost of goods. A $19.99 tier doesn't either once you add the regenerations users always do. To get margin you're either capping usage so hard the product feels stingy, or you're pricing at $49+ and competing with people who'll happily use Google's own Flow or Gemini app directly for less friction. The wrapper economics that worked for text (where a query costs fractions of a cent and you can give away a generous free tier) do not transfer to video. Video is metered in dollars per clip, not cents per thousand queries, and there's no free tier to subsidize acquisition.
That's the trap. The mental model a lot of us built during the cheap-text era ("inference keeps getting cheaper, give it away to grow, monetize later") is exactly wrong for video output. You are reselling a commodity at a per-unit cost you don't control, priced by a vendor who also sells directly to your customer.
Where a solo operator can still win
None of this means video is off-limits. It means raw generation resale is the wrong layer to compete on. The margin lives above it.
The defensible product is the workflow, not the generation. Editing, versioning, brand consistency, templating, the boring assembly work of turning raw clips into something a business can actually publish on a schedule: that's where you add value a customer will pay for repeatedly, and where your cost is your own labor and code rather than a metered per-second API. Gemini Omni Flash's editing capability is the interesting half here, not the generation half: a tool that lets a small business tweak and iterate a video without regenerating from scratch is worth more, and costs you less to run, than one that burns a fresh dollar on every attempt.
The other winning shape is high-value, low-volume. A product that generates a handful of expensive clips for a customer whose deliverable is worth hundreds (a real estate listing video, a product ad, a course intro) can absorb a few dollars of generation cost trivially, because the clip isn't the product, the outcome is. That's the inverse of the consumer-wrapper math: fewer generations, each attached to real revenue.
What I'd actually do before building
Run the unit economics before you write a line of code. Take your honest estimate of generations per user per month, multiply by clip length, multiply by $0.10 a second, add a regeneration factor of at least 1.5x because users never accept the first take, and add the $1.50/1M input for your reference material. That's your cost of goods per user. Then ask whether the price you can realistically charge clears it with room for everything else. If the answer is no at consumer price points (and for pure generation resale it almost always is), you don't have a pricing problem you can fix later, you have a business-model problem you need to fix now.
And price against the direct competitor honestly. Your customer can open the Gemini app or Flow and generate clips themselves. Whatever you build has to be worth more than that friction differential. "Same thing, wrapped" isn't, once the underlying cost is a dollar a clip that you both pay.
The honest take
I might be too pessimistic on one axis: prices at the frontier fall, and video won't be $0.10 a second forever. Text output dropped by orders of magnitude in two years, and there's no physical reason video won't follow. If you believe that curve bends fast, a consumer video product that's underwater today could be profitable in a year on falling costs alone.
But I wouldn't bet a business on a price cut that hasn't happened. Right now, today, video generation is metered in dollars per clip with no free tier, sold by a vendor who's also your competitor, and the cheap-text playbook does not port. Build on the layer you control (the workflow, the editing, the outcome) and let generation be a cost line you keep small, not the product you resell. If your idea only works when video gets cheap, you don't have a product yet. You have a bet on Google's price sheet.
Author
Lukas
@lukcombinator