How to upscale a low-resolution image without losing quality

What resolution and DPI actually mean

An image file is made of a fixed grid of pixels, say 1200 by 800. Resolution just describes that pixel count; DPI (dots per inch) describes how densely those pixels are packed when the image is printed or displayed at a given physical size. The same 1200-pixel-wide file can look sharp on a phone screen and soft on a large poster, because the number of pixels stays fixed while the physical size changes.

Why plain resizing makes an image blurry

When you drag an image larger in an editor, the software has to invent new pixels to fill the extra space, since the original file doesn't contain them. It does this through interpolation, essentially averaging the colours of nearby existing pixels. That's fine for small increases, but it produces the soft, slightly smeared look you get from stretching a small image a long way, because it can only guess at detail that was never captured in the first place.

When 2x makes sense and when you need 4x

A 2x enlargement (doubling both width and height) is usually enough to take a web-sized image up to a modest print or a larger on-screen banner, and interpolation alone can sometimes get away with it if the source was reasonably sharp. A 4x enlargement, taking a small image up to something print-sized or suitable for a large banner, is where interpolation clearly falls apart and a proper detail-reconstruction approach starts to matter.

What AI upscaling does differently

Rather than simply averaging nearby pixels, AI upscaling models are trained on large numbers of image pairs to recognise what plausible fine detail, such as fabric weave, skin texture or edges, tends to look like, and they reconstruct that detail at the new size instead of just smoothing it. The result is noticeably sharper than a plain resize, particularly at 4x.

The limits worth knowing

AI upscaling reconstructs plausible detail; it doesn't recover information that was never there. It can sharpen textures and edges convincingly, but it can't reliably invent small text so that it becomes newly legible, and it can occasionally alter fine facial detail in ways that don't match the original person. Treat it as detail enhancement, not a substitute for a proper source image.

Preparing the source image

• Always start from the largest and least compressed version of the image you have access to
• Avoid upscaling a screenshot of a screenshot; each generation loses more detail and compresses further
• If the source has obvious JPEG blockiness, a light cleanup pass before upscaling gives a noticeably better result than upscaling the artefacts along with the image

A rough guide to print sizing

As a working rule, a 1200-pixel-wide image prints at a genuinely sharp quality at around 10 centimetres wide at 300 dpi, the density typically used for good-quality print. Double the pixel width and you roughly double the sharp print width, which is the practical reason to upscale before sending an image to print rather than just stretching it in a page layout tool.

Typical situations where this comes up

• Turning a small logo or photo into a LinkedIn banner without it looking soft at full width
• Enlarging a product photo supplied by a manufacturer that's smaller than your listing template needs
• Bringing an old scanned photo up to a size where it can be printed or shared without looking pixelated

Try your image editor's free tools first

For small increases in size, it's worth trying your existing editor before reaching for anything else. Photoshop's 'Preserve Details' resample mode, GIMP's cubic interpolation, or similar options in free tools can handle modest enlargements reasonably well and cost nothing.

A note on video and animated content

Everything above applies to still images. Enlarging video frames introduces the same interpolation softness, and consistency across frames adds an extra layer of difficulty, since a model has to keep reconstructed detail stable from one frame to the next rather than treating each frame in isolation.

Checking the result before you use it

Whichever method you use, zoom in on the enlarged image before publishing or printing it. Look closely at edges, text and any faces for artefacts or odd smoothing, and compare against the original at the same zoom level so you're judging the upscale fairly rather than against a memory of how sharp the small version looked on screen.

Using an AI upscaler

Pitanga Labs' AI Image Upscaler enlarges an image by 2x or 4x with detail reconstruction, for 2 credits per image, and accepts PNG, JPEG or WebP files up to 10 MB, which covers most everyday upscaling jobs.