Eight times bigger

Above the picture there is a line saying what just happened — which pass ran, in what order, how many times. That line exists because “8×” on its own is a marketing claim and not a description. With Lanczos or Mitchell it is a single resampling pass straight to the final size, which is mathematically the right way to do it: chaining two 4× passes would filter the first pass’s own errors a second time. With edge-guided it is three doublings, because that method works by filling the holes in a doubled grid and has nothing else to fill.

Sixty-three of every sixty-four pixels in an 8× result were computed. What you are looking at is a very careful account of what the remaining one pixel implies — and on a 200-pixel source, that is a picture of an inference rather than a picture of a thing.

Fit-to-screen will not tell you, because the screen is shrinking the result back down and hiding exactly the difference you are trying to judge. Press 1:1, move the loupe to the most detailed part of the picture — a hairline, a brick course, a label — and hold the compare button. If the plain stretch and the result look the same at that magnification, the source had nothing left to give and the factor is not the problem.

If the result looks crunchy rather than soft, try Mitchell. If diagonals look like stairs, try edge-guided. If the whole frame looks like smooth plastic, the clean-up pass is too strong — drop it to Light or turn it off, because at 8× every bit of smoothing you apply first is smoothing you see eight times over.

Is 8× ever the right answer?

Yes, in two cases. A poster seen from two metres needs far less detail per inch than a print held at arm’s length, so a small source blown up hugely can look perfectly good on a wall. And pixel art genuinely wants 8× nearest-neighbour, because there the squares are the artwork. Outside those, 8× is usually two steps past where the picture stopped having anything to say.

Why is the 8× chip greyed out?

Because eight times your picture is more pixels than a browser canvas will hand back on this device. The tool works out the largest factor that still fits and says so under the picture; on a phone that limit arrives a lot sooner than on a laptop.

Would a paid tool do better at 8×?

At 8× a trained model will give you something that looks sharper, because it is inventing plausible texture rather than spreading the texture you have. Whether that is better depends entirely on whether you need the picture to look good or to be true. For a wall print, invented texture is fine. For anything where somebody might identify a detail, it is worse than a blur, because a blur does not look confident.

• Four times overThe usual answer when a picture is close but not close enough. Opens at 4×.
• Double the sizeThe factor with the least to apologise for. Opens at 2× on a sharp source.
• What enlarging can and cannot get backWorked through at 2×, 4× and 8×, with the part that is a guess named as one.
• Enlarged and still softSix reasons a bigger picture looks worse, and which one you actually have.