Google Just Dropped AI Image Generation to 3.4 Cents. If Your Product's Moat Is 'It Makes Images,' That Moat Just Repriced.
Google Just Dropped AI Image Generation to 3.4 Cents. If Your Product's Moat Is 'It Makes Images,' That Moat Just Repriced.
On June 30, Google shipped Nano Banana 2 Lite. It turns a text prompt into a 1K-resolution image in about four seconds, at $0.034 per image, roughly five times faster than the standard Nano Banana model, built explicitly for high throughput and scale. Alongside it, Gemini Omni Flash came to developers at $0.10 per second of video output, matching Veo 3.1 Fast's rate. Both are live in AI Studio and the Gemini API today.
The tech press covered this as a speed-and-cost story, which it is. But if any part of your product's value proposition is "it generates images" or "it generates video," this is a pricing-page story, and it's about your pricing page, not Google's.
Three and a half cents is a floor, and floors move markets
Put the number in context. A few years ago, generating a usable image from a prompt was a differentiator you could build a company on. Then it was a feature. Now it's a line item that costs less than the payment processing fee on the transaction it's part of. Video followed the same curve and is now measured in dimes per second.
When the raw capability gets this cheap, two things happen at once, and they pull in opposite directions. The cost of shipping an image or video feature collapses, so more products add one. That grows the market. And the price a customer will pay for "I can generate an image" collapses too, because they can now do it themselves for pennies. That shrinks the margin on anyone selling the raw capability with a markup.
Which of those two forces hits you depends entirely on what you're actually selling. And a lot of solo operators are, whether they admit it or not, selling the raw capability.
The thin wrapper just got squeezed from both sides
Here's the setup that's now in trouble. You built a product that takes a user's text, calls an image model, applies a bit of UI, and charges a subscription. The pitch was convenience: they don't have to touch an API, you handle it, they pay you monthly.
That business is getting compressed from below and from the side simultaneously. From below, because your cost of goods was already the API call, and while cheaper input helps your margin on paper, it also means the thing you're marking up is now nearly free, which makes your markup obvious and your price hard to defend. From the side, because a competitor can now undercut you overnight using the same cheap model, and the next person can undercut them, all the way down to "basically the API price plus a thin skin." When the underlying commodity gets cheap enough, "I'll call the API for you" stops being a business and becomes a race to zero with better landing pages.
I've watched this happen to a small thing I built. It was a nice generator with a nice UI, and it made real money for a while. It also had no floor, because the day the model got cheaper and more capable, my entire reason to exist was "I have a login screen and you don't have to read the docs." That's not a moat. That's a speed bump.
Where the value actually moved
The value didn't disappear. It moved up a layer, to the parts of the problem the cheap model doesn't solve.
Think about what a customer generating images for their business actually needs beyond the image. They need brand consistency: the same style, palette, and feel across fifty assets, not fifty one-off generations that all look slightly different. They need it to fit a specific workflow: dropped into their product listings, their ad templates, their social calendar, in the exact dimensions and formats they use, without manual fiddling. They need it to be reliable enough to run unattended and predictable enough to plan around. They need the taste to know which of the twelve generations is actually good.
None of that is the generation call. All of it is the wrapper, in the good sense of the word: the workflow, the constraints, the domain fit. The cheap model made the expensive part free and left the scarce part exactly where it was: understanding one customer's specific job well enough to deliver a result they'd have paid a designer for, at a price and speed a designer can't match. That layer got more valuable when generation got cheaper, not less, because now the generation isn't the cost or the bottleneck. Your judgment is.
What I'd actually do
If you're building on image or video generation, stop pricing on generation. Per-image and per-second pricing that mirrors your own API cost is just volunteering to be arbitraged the next time Google or anyone else cuts the underlying rate, which, on current trend, is roughly every quarter. Price on the outcome the customer came for: the finished campaign, the consistent product-shot set, the ready-to-post video, the thing that saved them an afternoon or a freelancer invoice.
Then spend your build time on the parts the model can't commoditize. Brand and style consistency across a batch. Tight integration into wherever the customer actually works. Templates that encode taste so the non-designer gets a good result on the first try. The eval loop that catches the bad generations before the customer sees them. That's the asset, and it compounds while the token and pixel prices keep falling underneath it.
And keep the model swappable. Today Nano Banana 2 Lite is the cheap fast option; in three months it'll be something else at a lower price. If your product is genuinely the outcome, the generator underneath is a supplier you replace when a cheaper one shows up, which means every price cut like this one is a tailwind for your margin instead of a threat to your existence.
Here's the honest counter-take. Cheap generation is genuinely good news, and the doom framing is wrong. A lower floor means more people can afford to ship media features, which means a bigger market, which means more room for a well-positioned solo operator, as long as you're selling the outcome and not the API passthrough. The three-and-a-half-cent image doesn't kill the opportunity. It kills the version of the opportunity where your only edge was standing between the customer and the API. If that was the plan, this week is your notice to move up a layer.
Author
Lukas
@lukcombinator