The fight over AI images is being fought, increasingly, in cents. On Monday Google released an update to its image model, Nano Banana 2.1, and the news that mattered most was not a new capability but a new price: $0.0336 to generate a standard 1K-resolution picture, down from $0.067 the week before.
The cut, roughly 50 percent, came quietly, folded into an update to the company’s pricing page. The new model is built on Gemini 3.6 Flash, the latest version of Google’s lightweight image engine, and the company said it would roll out across the Gemini app, Google AI Studio, Search’s AI Mode and its advertising and enterprise tools.
The model’s abilities did expand. It accepts up to 14 reference images in a single prompt, letting a developer feed it examples of characters and objects and ask it to keep them consistent across a conversation. Google said the update improves multi-turn editing, text rendering and masked local edits, where a user can rework part of an image without regenerating the whole thing.
The price cut is the sharper signal. At $30 per million image tokens, Nano Banana 2.1 is now Google’s cheapest image API, and the company is moving developers off the older model quickly. It said the previous Flash image model, Nano Banana 2, would be retired on October 29, giving developers under four weeks to migrate. A deadline that short leaves little room to shop around.
Google has leaned on a deliberately silly name for the series. The “Nano Banana” label, an in-house nickname that escaped to the public, now fronts the company’s image work, and the naming has stuck in a way the technical designations never did. The update keeps the name and swaps the engine underneath.
The model runs at resolutions from 1K up to 4K, with a context window Google says reaches a million tokens, enough to hold a long back-and-forth of edits and reference images in a single session. On the Arena image-editing leaderboards that the AI community watches closely, the update has landed near the top, a ranking Google was quick to circulate.
The economics are stark at the low end. A thousand standard images now cost about $33.60 before other charges, and batch processing halves that again. For the small studios and solo developers who drive much of the image market’s volume, the number is the difference between experimenting and passing.
Character consistency is the gap Google is chasing. Earlier image models could produce a striking single frame but drifted when asked to render the same face in a second scene, which made them awkward for comics, ads and games. Holding a cast steady across a conversation is the feature studios actually pay for, and it is the one Google chose to advertise alongside the price cut.
The speed of the move is the point. Google has spent the past year jockeying with OpenAI and others for the image generation market, and the pattern has been the same across the industry: a new model lands with a splash, then the price starts dropping. Nano Banana 2.1 shortens the cycle, cutting cost at the same moment it adds capacity.
The same day, Google made a quieter announcement that points in a different direction. It opened Playground, an experimental platform built with Unity, the game engine company, that turns a text prompt into a playable game. A user describes a game, and the system assembles a working version from that description.
The two moves fit together. Nano Banana 2.1 drives the cost of making images down toward zero; Playground drives the cost of making games toward the same floor. Google’s argument, analysts said, is that content creation is becoming a task anyone can hand to a model, and the company wants to own the tools people reach for first.
The game announcement is early-stage. Unity and Google described Playground as experimental, and the games it produces are, for now, simple. But the direction matters to an industry where the cost of producing a game has climbed for years while the tools have stayed roughly the same. A prompt-to-game pipeline, if it matures, would do to game studios something like what text-to-image did to stock photography.
Google is not alone in cutting prices. OpenAI has trimmed its own image pricing through the year, and the market for generated images has settled into a race to the bottom that favors whoever has the cheapest inference and the biggest distribution. Google’s advantage is the distribution: the model ships inside Search, Android, YouTube and its advertising network, where volume can absorb margins that rivals cannot match.
The next few weeks will show how developers respond. The migration deadline forces a decision, and the half-price image is the incentive to choose Google. Whether rivals answer with their own cuts is the question the market is now waiting on, and the answer will shape what it costs to make a picture, and eventually a game, for the rest of the decade.
