Upscale images 4× with AI

Most low-resolution images were not taken that way. A website served a reduced copy so it would load faster. A chat app shrank the photo before sending it. Someone cropped a distant face out of a group shot, or the only copy left came from an old phone or an old social media album.

Once those pixels are gone, a normal resize cannot bring them back. Bicubic or bilinear resizing, the kind built into every editor, spreads the existing pixels over a larger area and blends between them. The image gets bigger and evenly soft, and any blockiness in the original grows with it.

Upscale 4× is a super-resolution network followed by a detail model. The first network predicts the four-times-larger image that best fits your pixels. Where the small image clearly contains an edge, the result gets a clean, continuous edge instead of a staircase. The second model adds the fine texture a sharp photo has at that size, such as fur, foliage, stone and fabric, guided by the colours and shapes that are really there.

Both were trained on pairs of a high-resolution photograph and a small, realistically damaged copy, including the compression and light noise small images usually carry. So the model copes with a little JPEG damage while it enlarges. For a photo that is very noisy or heavily compressed, run Enhance first and upscale the clean result.

It is made for photos of nature, people, places and objects. Small text in the photo stays as soft as in the original: the model protects text it detects and does not guess letters, but small, dense print can still come out wrong. For documents and screenshots, upscaling is not the right fix.

Width and height are multiplied by four, so the number of pixels grows sixteen times. A 1000 × 750 image becomes 4000 × 3000 (12 MP).

The output is limited to 80 MP, so Upscale 4× accepts images up to 5 MP, for example 2560 × 1920. The free plan accepts files up to 3 MB; paid plans up to 20 MB. A photo larger than 5 MP often has enough pixels already: check your target size, or crop it before upscaling.

For prints viewed up close, 300 pixels per inch is the usual target. Posters seen from a few steps away look fine at 150 to 200. Divide the pixel size by the pixels per inch you want to get the print size. A few examples after a 4× upscale, at 300 pixels per inch:

• 800 × 600 becomes 3200 × 2400, enough for about 10.7 × 8 inches (27 × 20 cm).
• 1200 × 900 becomes 4800 × 3600, enough for about 16 × 12 inches (41 × 30 cm).
• 1080 × 1080, a typical social image, becomes 4320 × 4320, about 14.4 inches (37 cm) square.
• The print-size calculator works it out for your exact file.

Use 4× for most jobs: product photos for zoom, images for slides and screens, and prints up to roughly A3 from an original of about 1 MP. It changes the picture the least and is the safer choice for faces and products. Use Upscale 8× when the original is a thumbnail of a few hundred pixels, or when you need a poster from a small landscape or architecture photo. 8× uses 2 credits per photo.

The quality of the result depends more on the input than on anything else:

• Start from the largest, least-compressed copy you can find. The original from a camera roll beats a version saved from a chat.
• Clean heavy noise or compression first with Enhance, then upscale.
• Crop after upscaling, not before, if you only need part of the picture.
• Check faces, small text and fine patterns at 100 % zoom before you print or publish.
• Save the result as PNG or a high-quality JPEG so you do not add new compression on export.