Enlarging a screenshot, and what it will not fix
Screenshots enlarge well everywhere except the small text, and recapturing beats enlarging whenever you can.
Steps
- Try to recapture before enlarging
Zoom the page or the app first and take the screenshot again, since a fresh capture has real pixels where an enlargement only has guesses. - Use Illustration mode for interface captures
Flat colour and hard edges behave like line art, so the illustration model suits a UI screenshot better than the photo model. - Crop to what you need before enlarging
Cutting the picture down to the region you actually want keeps the output smaller and the processing shorter. - Judge the text at 100%
Compare before and after at full size and look at the smallest text, which is where an enlargement fails first.
Screenshots respond well to enlargement everywhere except one place: small text. That is unfortunate, because small text is usually the reason someone wants the screenshot enlarged. Everything else — panels, icons, charts, photographs inside the capture — comes back clean. Open Pixmend if the text is not the point, and read the rest if it is.
Recapture instead, if you still can
A screenshot is not a photograph. The pixels were drawn by software, so the same picture can usually be produced again at a larger size, with real detail rather than inferred detail. That is always the better result.
- A web page: zoom the browser in before capturing, or use a device-pixel-ratio setting in the developer tools to capture at twice the size.
- An app window: make the window bigger, or raise the display scaling, and capture again.
- A document, slide, or chart: export it as an image or PDF at the size you need rather than photographing the screen showing it.
Enlargement is for screenshots you cannot take again: something from a chat thread, a bug report, an archived page, a video frame. If you can retake it, retake it. No model competes with a capture that has the pixels in the first place.
What enlarges well, and what does not
| Part of the capture | Result | Why |
|---|---|---|
| Photographs and video frames inside the shot | Good | This is what the models are for |
| Icons, logos, flat panels | Good | Hard edges against flat colour are easy to reconstruct |
| Charts, diagrams, large headings | Usually good | Big shapes with clean boundaries |
| Body text at a readable size | Acceptable | Edges get cleaner, letterforms stay recognisable |
| Small or anti-aliased text | Poor | The model has no notion of letters, only edges |
| Hairlines, one-pixel borders, dithered gradients | Mixed | Thin lines can thicken or shift slightly |
Small text is the honest failure. The model does not know what a letter is; it reconstructs plausible edges from the pixels it is given. When a character is only a few pixels tall, the pixels no longer describe that character uniquely, so what comes back is a shape that looks like crisp text at a glance and is not the same text under inspection. If the point is legibility — a serial number, an error message, a value in a table — enlargement will not deliver it, and a result that looks sharp is more dangerous than one that looks blurry.
Settings that suit a screenshot
Use Illustration mode for captures of an interface. A UI is flat colour bounded by deliberate edges, which is the same structure as line art, and the photo model tends to soften those edges. Switch to Photo when the capture is mostly a photograph or a video frame with a small amount of interface around it.
Denoise only applies in Photo mode. A screenshot saved by the operating system is lossless and has nothing to remove, so leave the setting at its default. A screenshot that has been through a chat app is a JPEG with compression damage around every edge, and if you are running it in Photo mode, a stronger setting cleans that up as it enlarges.
Crop before you enlarge. Screenshots are mostly empty interface, and cutting down to the region you actually want gives a smaller output for the same result. The input limit is 12.6MP, which a single full-screen capture stays inside, though a very wide multi-monitor capture may not.
Export as PNG. A screenshot is flat colour and text, which is exactly the content JPEG handles worst, and saving the result as JPEG re-introduces the artefacts you just removed.
A sensible order of attempts
- Recapture at a larger size. Faster and better than anything below it.
- If you cannot, crop and enlarge in Illustration mode, and check the result at 100%.
- If the small text still matters, read it in the original and type it out. Reproducing the words is honest; a sharpened guess at them is not.
That third step is worth stating plainly because it is what the tool cannot do for you. Nothing here invents detail — the first run downloads about 11.6MB of models and runtime, then everything happens on your own GPU inside the browser, and what comes back is what the file already contained, at a larger size. For a screenshot, that fixes the picture and not the letters.