Pixels, Dimensions, and What Resizing Changes
A digital image is a grid of pixels, and resizing rebuilds that grid at a new size. Shrinking merges several source pixels into each new one, which is a safe operation, the result stays sharp because information is being condensed, not invented. Enlarging is the opposite: the resizer must fill pixels that never existed, and plain interpolation can only produce them by averaging, which reads as blur.
That asymmetry is the single most useful thing to know. Downscaling is essentially free in quality terms, upscaling is not. When you genuinely need a larger image, plain resizing is the wrong tool, the AI Image Upscaler reconstructs detail with AI instead of averaging and produces a dramatically sharper enlargement.
This resizer also steps large reductions down in halves internally rather than jumping in one leap. Single-jump downscaling past half size skips over pixels and creates subtle jagged patterns, stepped scaling averages every source pixel properly and keeps fine detail smooth.