A 500x500 pixel image has exactly 250,000 pixels of information, no more and no less. Resizing it up to 2000x2000 doesn't add real detail, because that detail was never captured in the first place, and the mathematics of resizing has to fill in 16 times as many pixels using only the original data as a source.
How a standard resize actually fills in the gaps
Standard resizing algorithms, like bilinear or bicubic interpolation, estimate each new pixel by blending the values of nearby original pixels. This works reasonably well for small enlargements, but the more you scale up, the more pixels are essentially educated averages of a shrinking pool of real information, which is exactly what produces that soft, blurry, or blocky look at higher magnifications. There's no way around this with interpolation alone: it can only interpolate between data that exists, it can't invent new detail.
What AI upscaling actually does differently
AI upscaling tools use a model trained on huge numbers of low-resolution and high-resolution image pairs, learning what plausible fine detail (hair strands, fabric texture, sharp edges) typically looks like when a similar low-resolution pattern is enlarged. Rather than blending existing pixels, the model generates new pixel values based on patterns it learned, effectively hallucinating detail that's statistically likely to belong there rather than recovering detail that was actually captured. AI Upscale Image applies this approach, and it noticeably outperforms a standard resize on photos with faces, text, or fine texture, though it can occasionally invent detail that wasn't in the original scene.
For genuinely sharpening detail that's present but soft (as opposed to enlarging), Image Sharpener is a better first step, and Image Noise Reducer can clean up grain before upscaling amplifies it.
Why starting resolution matters more than any tool
No upscaling method, AI or otherwise, produces results as good as capturing the image at a higher resolution in the first place. If you have a choice, always keep the largest original file available and downscale for specific uses rather than starting small and upscaling later, since downscaling (throwing away detail you don't need) is a far more forgiving operation than upscaling (inventing detail you never had). Resize Image handles the straightforward downscale case for everyday resizing needs.