A plain resize interpolates between existing pixels, which smears fine detail. Real-ESRGAN models what the missing high-frequency detail should look like at the target resolution — skin texture, fabric weave, hair edges — so the larger image reads sharper, not softer.

For photos with decent source detail, 2× gives the cleanest rebuild. For very small sources or the largest possible output, 4×. UI screenshots and documents sharpen legibly at 2×; pixel art prefers 2× to avoid over-smoothing.