How far can a photograph be enlarged before it stops being itself?

Roughly 1.3 to 1.75 times is the range where enlarging a photograph recovers detail that is genuinely there. Past about 3.5 times, the process is no longer recovering anything — it is predicting what should plausibly be present.

Products here are sized to stay in the first range rather than pushed into the second to reach a bigger number on a price list.

The two things that get called the same word

"Upscaling" covers two different operations and the distinction is the whole subject.

Reconstruction takes edges and textures that exist in a degraded form and firms them up. The information was there; it was just spread thin. This is what happens inside the faithful range, and the result looks like your photograph.

Prediction fills space with what a model expects. A pattern appears on a dress that was never that pattern. A face gains detail the negative never held. This is what happens past the ceiling, and the result is attractive and not yours.

There is a hard technical edge too

The reconstruction model in common use refuses input above roughly 2.1 megapixels outright — it will not accept a very large source at all. That is a useful reminder that these tools have operating ranges rather than infinite headroom, and that a bigger input is not automatically a better one.

For the modest enlargements a well-sized print needs, conventional resampling is already excellent and does not invent anything.

Why the range is narrow rather than a single number

Enlargement is not one operation applied uniformly. How far a given photograph travels depends on what it is a photograph of.

Flat areas of even colour scale a long way — sky, a painted wall, fabric with no weave. Fine repeating texture scales badly, because reconstruction has to decide where each element sits and a small error repeated across a jumper is visible. Faces sit in the middle and matter most, which is why the working range is set by faces rather than by the easiest part of the frame.

What this means when you are choosing

The size question is really a question about your source. A sharp photograph from a recent camera has room to go large. A small scan of a damaged print does not, and asking it to fill a large frame means accepting invented detail in somebody's face.

If you want the larger piece, the answer is a better source rather than more processing.

Related

Will it still look like them covers why faces in particular are left alone, and what resolution does a photo need covers the floor at the other end.