Cheap Pixels Are Here. Now Decide What They're For.

The price of making a generated image just fell through the floor. So did the friction of turning a meeting into notes, a keyboard tap into an action, a podcast into a newsletter. A lot of things you used to scope as a project now cost less than a coffee.

That sounds like good news, and mostly it is. But cheap does not mean free, and it does not mean it belongs in your product. Let me catch you up on what shipped and where you should actually put it.

Drafting just got absurdly cheap

Google released Nano Banana 2 Lite at $0.034 per 1,000 images, with output in four seconds. That is drafting money. You can generate thousands of rough visuals and throw most of them away without a finance conversation. Its sibling, Gemini Omni Flash, does video editing from plain-language prompts at $0.10 per second of output.

Apple pushed the same idea into tools people already own. Apple Creator Studio added Auto Mask that reads skin, hair, and sky with no manual tracking, plus image generation inside Pixelmator, Keynote, and Pages. The pattern is clear: fast, cheap generation is now a feature of the software your team already opens.

Speed is the tradeoff, not a free lunch

Cheap and fast come with a bill. Omni Flash caps videos at 10 seconds, and Google itself admits character consistency breaks when scenes change or the camera pans. Lite keeps decent prompt adherence, but it is built for volume, not the final frame.

So split the job. Use Lite to draft and explore, then move to a heavier model when accuracy matters more than speed. Google structures its own lineup this way, from Lite up to Nano Banana Pro for work where getting it right beats getting it now. If your team treats a cheap draft as a finished asset, you will ship the seams.

The line between your workflow and your product

This is the call that matters. A cheap generator can speed up how your team works, or it can become a feature you ship to customers. Those are different bets with different stakes.

Inside your workflow, the risk is low. Gemini in Sheets can build a tracker or write a formula from a plain-English prompt, and if it's wrong, you fix it and move on. Pocket sold over 130,000 of its $129 note-taking pucks by making meetings easier to capture, a clean workflow win. Put these where a bad output costs a redo, not a customer.

When AI becomes the interface

Some of these plays are not tools, they are bets on where people spend attention. Acti put AI agents inside the smartphone keyboard, so a user can translate a message or drop a live stock price without leaving the chat. Founder Young Wang calls the keyboard "a context layer that genuinely belongs to the user instead of the platform."

Riverside made a smaller, smarter bet. Instead of building a blank-page newsletter tool to fight Substack, it turns recordings users already made into newsletters. CEO Nadav Keyson said starting from a blank page is the wrong ask when "the ideas are already there, in the conversation." That is the tell: the best product bets add AI where your users already have context, not where you wish they did.

The deep cut

Cheap generation is a workflow decision by default and a product decision only on purpose. The failure mode is drifting from one to the other without noticing. You start using Lite to mock up screens, it looks good, and someone ships those mockups as real UI to customers. Now the 4-second draft with wobbly character consistency is your brand.

Before any of this touches a customer, name it. Workflow tools need a fast-fix loop and low stakes. Product features need consistency, a fallback for when the model breaks, and a reason it beats sending the user to a chatbot. Write that line down before your next review, because the price tag makes it very easy to skip.

Three questions for your team

1. For each generative feature on our roadmap, is this speeding up how we work or is it something a customer will see? What breaks if we're wrong about which?
2. Where are we treating a cheap draft as a finished asset, and what's our plan for the 10-second cap and the consistency gaps?
3. Which AI feature earns its place because the user already has context there, and which one is just a chatbot we bolted onto our app?