Editing

Resizing vs resampling: the honest limits of upscaling

· ~6 min read · Last verified: 2026-11-16

Your file is 2400 pixels wide. The print you want "needs" 3600. Every image editor offers a dialog that will happily type 3600 into that box — and the internet is full of "convert to 300 DPI in three clicks" tutorials that quietly do exactly that. Here's what actually happens when you press the button, in the labs' own words, and what your real options are when the pixels run out.

Key takeaways

  • Resize = change the size label (inches/PPI) without touching pixels. Harmless, and usually unnecessary.
  • Resample = change the pixel count. Down: safe. Up: interpolation — new pixels are guesses.
  • WHCC: switching a file's tag from 72 ppi to 300 ppi "will not add any quality to the file".
  • Nations: resolution "cannot be added to an image once it is captured by enlarging the image, it can only be removed."
  • AI upscaling can look convincing at modest factors — WHCC mentions it as something to test, not a guarantee.

Resizing: changing the label, not the photo

Resizing (what Photoshop does with "Resample" unchecked) rewrites the metadata that pairs pixels with inches. A 2400 × 3000 file labelled 300 PPI is "an 8 × 10"; relabelled 150 PPI, it's "a 16 × 20". The pixels are untouched — identical file, different arithmetic. WHCC's file-prep help turns this into a diagnostic: uncheck Resample, type 300 ppi, and whatever size appears is "the optimum size for printing as long as the original file is clear". Relabelling is also the correct reply to a lab that "requires 300 DPI files": the requirement means pixels-per-inch at final size, not a magic tag (the vocabulary is unpacked in what DPI actually means).

Resampling: making up pixels

Resampling changes the pixel count. Downsampling (fewer pixels) discards information but does no harm — your 6000-px original becomes a fine 1080-px web image, and as the platform specs show, nothing beyond 1080 wide survives Instagram anyway. Upsampling is the dangerous direction: every new pixel is interpolated — a weighted average, or a neural network's confident guess, but a guess. Edges soften or acquire halos, textures turn waxy, and the effective information per inch does not improve even though the PPI arithmetic now says "300". Cambridge in Colour's enlargement tutorial is direct about the ceiling: prints enlarged past the native size "will become highly dependent on the quality of interpolation and sharpening."

The labs put it more bluntly. Nations Photo Lab's resolution guide: resolution "cannot be added to an image once it is captured by enlarging the image, it can only be removed." WHCC, on relabelling: "Changing the resolution on a file does not affect the image quality" — and converting 72 ppi to 300 ppi "will not add any quality to the file, the appearance will be the same." Both statements point the same way: the sensor's capture is the budget; software only spends it.

Where AI upscaling fits (honestly)

Modern AI upscalers are genuinely better than classic bicubic interpolation at inventing plausible detail — WHCC's help acknowledges they exist ("There are various programs available that will use AI to add quality to a file"), while framing them as something to evaluate, not trust. The honest framing: an AI upscaler adds plausible pixels, not true pixels. On smooth subjects (skies, skin, studio backgrounds) at factors up to about 2×, results can be convincing. On text, architecture, crowds, and anything with real fine texture, the model guesses — sometimes beautifully, sometimes into uncanny mush — and no amount of guessing restores information the sensor never recorded. A 2× AI upscale of a soft photo is a sharper-looking soft photo, not a 4× larger good photo.

What actually works when you're short

The pre-order sanity check

Before resampling anything, run the original pixels through the DPI calculator at your intended size. If the verdict band is already "good" or better, resampling gains you nothing. If it's below the lab minimums, no software button honestly fixes it — pick a smaller size, a more forgiving surface, or a farther wall.

Related reading