AI IMAGES

Draft fast, then commit.

Google is bringing from-scratch image creation into Search, powered by its fast, low-cost Nano Banana tier. For creators, that tier is a draft engine: rip twenty quick comps, then commit one to a real render.

TLDR

Google is bringing AI image creation into Search results, powered by its fast, low-cost Nano Banana tier. The creator move is to treat that speed as a draft engine, not a finisher.

•  Announced July 14 alongside Google Images turning twenty-five, it lets you type a prompt into your results and get a custom image built on the spot.

•  It runs on Nano Banana 2 Lite, Google's quick and inexpensive image tier, roughly four seconds an image and a few cents each.

•  The use for creators is a thumbnail sprint: generate twenty fast comps to find composition and framing before you commit one to a slow, finish-grade render.

•  It is announced, not fully live, rolling out over the coming weeks in English only, and the output is draft quality, so treat it as scaffolding rather than a final frame.

•••

Google is about to put image generation inside the most visited surface on the internet. Starting to roll out July 14, its Search results can build a custom image from a typed prompt, with no separate tool and no new tab. The image shows up where the question was asked.

For most people that is a curiosity. For anyone who makes images for a living, the telling detail is which model runs it: Nano Banana 2 Lite, Google's fast and inexpensive tier, not its slow finish-grade one. Speed and low cost are the whole point.

That combination changes what the tool is for. A model this quick is not where you make the final picture. It is where you throw twenty rough ones to find the one worth finishing.

WHAT CHANGED

Image-making comes to Search.

Google is adding from-scratch image creation directly inside its AI Overviews, the summaries at the top of Search. Type a description into your results and it generates the picture in place. Announced July 14, tied to Google Images turning twenty-five and a redesigned gallery. The largest surface on the internet is becoming a place you make images, not just find them.

Two things follow. Discovery shifts, because AI images now live inside Search itself. And a very fast image model just got pointed at an enormous audience, which tells you what kind of model it is: quick, light, and built for volume rather than for the final frame.

THE MODEL

Built for speed, not the finish.

The engine is Nano Banana 2 Lite, the quick and low-cost member of Google's image family, which shipped June 30 at roughly four seconds and a few cents an image. It is not trying to be the most detailed model in the room. It is trying to be the fastest useful one. That is exactly the tool you want early in a project, and not the one to judge by a single hero image.

Hold the quality claims loosely, as always. What matters here is not that this is the finest image tier. It is that a good-enough image tier is now nearly instant and nearly costless, which quietly changes the rhythm of how you work.

THE CRAFT

Twenty drafts before one render.

Here is the habit worth taking from it. When an image costs seconds and pennies, stop trying to nail the picture on the first prompt. Rip twenty fast comps to find the composition, then commit only the winner to a slow, finish-grade model. The fast tier scouts. The slow tier finishes.

This is how photographers and illustrators already think, contact sheets before the final print, thumbnails before the painting. Instant image models finally make that loop quick enough to run every time, on tools like this one and on any fast model you already have.

THE LENS

Change one variable each pass.

To make a sprint useful, vary one thing at a time. Hold the subject steady and move only the framing, or the light, or the palette, across a run of quick drafts. A sprint where everything changes at once teaches you nothing. A sprint that changes one variable shows you exactly what that variable does.

Keep the two or three comps that work and throw the rest away without ceremony. The point of a draft is to be discarded. You are buying information about composition, for pennies, before you spend real time on a render.

THE CATCH

Announced, not everywhere yet.

Keep the timeline honest. This is coming to Search, not fully shipped, arriving over the coming weeks and starting in English where Google's AI image mode already works. If you cannot find it yet, that is expected. The rollout is real but partial.

And the output is draft grade by design. Do not mistake a fast comp for a finished frame, or hand a client the quick version. The tier earns its place as a scout, not a closer, and that is the whole reason it is useful.

THE TAKEAWAY

Let the fast model scout.

The headline is images in Search. The habit underneath is bigger and portable: use your fastest, lowest-cost image model to explore, and your slowest to commit. That division holds no matter which tools you use, and it gets more valuable as fast models spread to places like Search.

So when this reaches you, do not judge it as a finisher. Point it at the messy front of your process, throw drafts at it, and keep the two that teach you where the picture wants to go. Then finish somewhere else.

•••

I am putting together a draft sprint: a compact prompt scaffold for firing off twenty fast image comps that vary just one thing at a time, framing, light, or palette, so you find a composition before you commit it to a slow, finish-grade render, with a worked example for a poster, a portrait, and a product shot.

Want it when it ships? Reply with send me the draft sprint and I will get it to you.

A QUESTION FOR YOU

Which decision do you still make on the first draft?

Reply with the choice you tend to lock too early, the framing, the palette, the angle. The sprint will be built to keep that decision open across a run of fast comps.

If this was useful, forward it to a creator who still tries to nail the final image on the first prompt.

Until next time,

Luxe Prompting

Luxe Prompting

AI Image Generation for Creators