Stockertim Dictionary · Artificial intelligence
Upscaling
Also called AI upscaling, image upscaling, super-resolution, upsampling
In short
Increasing the pixel resolution of an image, with AI adding plausible detail that it invents rather than recovering detail that was really there.
- Beginner
- Process
- Checked
In detail
Upscaling makes an image larger in pixels so that it can be shown or printed at a bigger size. Traditional methods stretch the existing pixels and smooth the result, which often looks soft. AI upscaling uses a model trained on pairs of low- and high-resolution images to add sharp edges, texture and fine detail that look natural.
The added detail is a prediction. The model guesses what a sharper version would probably look like, so it can invent features that were never in the original: the wrong lettering on a sign, extra texture in hair or a slightly changed face. For decorative images this is usually fine. For evidence, documents, product details or identifiable people, invented detail can be misleading.
Common factors are 2 times and 4 times on each side. Because both width and height grow, the number of pixels grows by the square of the factor, which affects file size and processing time. Designers check upscaled files at full size for artefacts before sending them to print.
In practice, upscaling rescues small generated images for large formats, prepares old artwork for print and enlarges product photos, with the original kept alongside.
Worked example
A 1,024 by 1,024 pixel generated image is needed for a printed poster at 300 pixels per inch.
- Upscale 4 times on each side: 1,024 x 4 = 4,096 pixels.
- Total pixels after upscaling: 4,096 x 4,096 = 16,777,216.
- Printed size at 300 pixels per inch: 4,096 / 300 = 13.65 inches on each side.
The upscaled file prints at about 13.65 inches square; the designer checks fine details for invented artefacts before approving it.
In practice
Designers upscale images for print and large screens, keep the original, inspect results closely and avoid upscaling where invented detail could mislead, such as photos of real events or documents.
Watch out
- Upscaling cannot recover real detail that was never captured; it creates a plausible guess.
- Doubling width and height multiplies the pixel count by four, not two.
Learn it properly
Official sources
- Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile · National Institute of Standards and Technology
- Images tutorial · W3C Web Accessibility Initiative
Facts last checked on 3 November 2026.
Education, not advice: the Stockertim Dictionary explains what terms mean. It never recommends buying or selling anything.