Enlarge Photos Without Turning Them Blurry

"I made it bigger and now it looks blurry." It's one of the most common complaints about enlarging images, and most of the time it isn't the enlarger's fault. Blur, softness and blockiness each have different causes, and once you can tell them apart, you can dodge nearly all of them.

Three problems that get blamed on upscaling

It helps to name what you're actually seeing:

• Softness: the whole image loses crispness because pixels have been spread thin. This is the natural cost of enlargement.
• Blockiness (pixelation): visible square steps along edges, usually from a crude, non-smoothed stretch.
• Existing blur: focus or motion blur that was baked into the original and simply becomes more obvious at a larger size.

Only the first two are influenced by how you upscale. The third was decided the moment the photo was taken.

Blur that was already there

No enlarger can undo a shot that missed focus or caught camera shake. Upscaling changes dimensions, not sharpness. If your source is blurry, enlarging it gives you a bigger blurry photo.

Before you enlarge, check the original at 100% zoom. If edges already look mushy at native size, no amount of upscaling will rescue them. Reach for a sharper source instead.

What a good upscaler actually does

Since I built the image upscaler on this site, let me explain what's genuinely happening under the hood, because "AI upscaling" gets used loosely.

There are two levels of quality in enlargement:

1. Smart resampling. Classical algorithms like Lanczos blend neighbouring pixels with a weighted mathematical filter instead of just copying them. This is what separates smooth curves from staircase edges, and it's the reliable baseline, our tool uses the pica library's Lanczos filter for plain resizing.
2. Neural super-resolution. For real enlargement, the tool runs ESRGAN, a neural network from the super-resolution research literature, via TensorFlow.js, directly in your browser. Instead of only blending existing pixels, the model was trained on millions of image pairs to predict plausible detail: how a sharp edge, a strand of hair or a brick texture should look at the larger size. That's why AI-upscaled results can look genuinely sharper than a classical resize, especially at 2x and 4x.

Two honest caveats. First, the model predicts plausible detail, not recovered truth. It cannot restore text or faces that were never resolved in the original. Second, because the network runs on your device, very large images take longer on modest hardware, the trade-off for your photo never leaving your machine.

Manage softness with the right factor

Softness scales with how far you push the enlargement. To keep it under control:

1. Prefer 2x over 4x whenever the target size allows. Doubling holds detail far better than quadrupling, even for the neural model, which has separate 2x and 4x variants for exactly this reason.
2. Match the factor to the display. If you only need the image 40% bigger, don't 4x it and then shrink, enlarge only as much as the layout truly requires.
3. Judge at real size. An image can look soft zoomed in at 400% yet look perfectly sharp at the size a viewer will actually see it.

A pre-flight checklist

Run through this before every enlargement:

• Is this the highest-resolution original I have? (Not a screenshot, not a messaging-app copy.)
• Is the original already sharp at 100%?
• Am I choosing the smallest factor that meets my size goal?
• Have I cropped and cleaned the image first, so I'm not enlarging waste? (Crop first, then upscale.)

After you upscale

A couple of finishing touches keep things looking their best:

• Choose the right format. PNG preserves every detail losslessly, WebP or JPG give a lighter file for the web with a barely noticeable trade-off.
• Don't over-compress afterwards. Enlarging and then crushing the result with heavy compression reintroduces the very artefacts you tried to avoid. If you need a smaller file, use a quality-controlled compressor and stay at moderate settings.

Blur isn't inevitable. Start sharp, enlarge with the right method for the job, pick a sensible factor, and judge the result at the size people will really see, and your enlarged photos will look clean, not fuzzy. For the print-vs-web question (how large is large enough?), see the companion piece: Upscaling images for print vs web.

Sources

• ESRGAN: Enhanced Super-Resolution Generative Adversarial Networks (Wang et al., 2018)
• Lanczos resampling (the filter behind high-quality classical resizing)