Your phone's gallery app can now remove a stranger from a holiday photo. Your smart display generates its own ambient artwork. The note-taking app you use for shopping lists offers to make a cover image for a recipe. None of these were headline features when you bought the device — they arrived quietly in updates, and most people use them without ever wondering how they work or what they cost the company to ship.

The answer to both questions is more interesting than the marketing suggests, and it explains why these features appeared everywhere at once.

Almost Nothing Happens on Your Device

Almost Nothing Happens on Your Device Lagenio

The common assumption is that a phone powerful enough to shoot 4K video is powerful enough to generate an image. In practice, only the lightest edits run locally — small object removal, background blur, basic upscaling. Anything that creates a genuinely new image is a network request to a data centre, which is why those features stall on a bad connection and why they sometimes stop working entirely when a manufacturer changes providers.

This matters practically for anyone comparing devices. A feature list that says "AI image generation" tells you almost nothing about the hardware. It tells you the manufacturer signed up with a model provider, and that arrangement can change at any point in the product's life. Several devices have quietly lost generation features mid-generation when a deal expired.

The Cost Structure That Made This Possible

Two years ago, a company wanting to offer image generation had to host models itself. That meant hardware, expertise, and a capital decision — which is why only large platforms offered it.

Metered access changed the shape of that decision entirely. Generation is now billed per image at fractions of a cent to a few tens of cents depending on resolution and quality, with no commitment and no minimum. Published rates for the Nano Banana API and competing image models sit openly on aggregation platforms that expose several models through one account — which is how a three-person app team ships the same capability as a hardware giant, and why the feature spread across the app stores in about eighteen months.

It also explains a pattern you may have noticed: free tiers that are generous at first and quietly tighten later. The company is paying per image. When usage grows, the economics change, and the limit appears.

Why Your First Attempt Is Never the Good One

Generate AI Images From Text In Seconds Unsplash

Here is the part that explains the user experience better than any feature list. Nobody accepts the first generated image — not you, not the developers testing the feature. Real usage runs three to eight attempts before something is worth keeping.

That multiplier is why apps push you toward a "quick" mode before offering the high-quality one. The cheap draft lets you find the direction you want; the expensive final render happens once. Apps that skip this step burn through their own margins, which is usually why their free allowance is stingier than a competitor's.

Knowing this makes you better at using these tools. Iterate on the cheap setting until the composition is right, then switch to high quality for the one you are keeping. It is exactly what the developers do internally.

Acer 27in FHD 1920x1080 IPS 120Hz Gaming Monitor | Office KB272 G0bi

List Price: $149.99

Where These Features Genuinely Help

The honest assessment after two years of these tools shipping everywhere: they are good at things that do not need to be true, and unreliable at things that do.

Backgrounds, textures, ambient artwork, concept images, and removing an unwanted object from a photo — these work well and save real time. Anything requiring legible text inside the image still fails regularly across every model on the market, which is why the poster-maker feature in your design app produces beautiful layouts with gibberish words. And anything depicting a real product or a real person deserves scepticism, both from you as a user and from the companies deciding what to publish.

What This Means for Buyers

Treat "AI image generation" on a spec sheet the way you would treat a bundled cloud service rather than a hardware capability, because that is what it is. Ask what happens when the free allowance runs out. Check whether the feature works offline — usually it does not. And bear in mind that a feature dependent on a third-party provider can change or disappear in a software update, which has already happened to several products.

None of that makes the features less useful. It just means the interesting question about a device is no longer whether it has AI image generation. Practically everything does now. The question is what the company is paying per image, and how long they intend to absorb it.