What a screenshot enhancer can and cannot fix
Most screenshots that look bad were fine when they were taken. The damage comes afterwards: the capture is scaled down to fit a message, re-encoded as a JPG, forwarded and compressed again. What arrives has soft text, blocky squares in flat backgrounds and grey halos around dark letters. A screenshot enhancer fills in pixels from learned patterns, tightens edges and takes out some of the blur, blocking and noise, which is often enough to make a softened capture usable in a document or slide again.
What it cannot do is read. If a word or number is an unreadable smear in the source, there is nothing to recover, and the model may produce characters that look plausible but are wrong: a 3 that turns into an 8, a comma that becomes a full stop. Nor can it redraw parts of an interface that were cropped off or covered.
Text legibility: what helps and what cannot come back
A quick test tells you what the screenshot enhancer can do. Open the source at 100% zoom: if you can read every character, even with effort, the enhanced version will be easier on the eyes, especially when shown large on a projector or a high-resolution screen. Dark text on a light background responds best; thin fonts, tiny labels and text laid over busy images give the model the least to work with.
If you cannot read a word in the source, treat whatever appears in the output as decoration rather than information. That matters because people screenshot things in order to keep them: order numbers, amounts, dates, codes, file paths, names and URLs. Check each of these character by character against the original, and never use the output of a screenshot enhancer as proof of text you could not read before.
Crisp edges on buttons, icons and charts
Interfaces are built from one-pixel lines: borders, dividers, focus rings, table rules and chart gridlines. Ordinary resizing stretches each of them into a soft grey band two or three pixels wide, which is why enlarged screenshots look blurry even when the original was sharp. A screenshot enhancer sharpens those edges, so buttons, icons, rounded corners and checkboxes keep defined outlines at the larger size.
Under the hood, the screenshot enhancer is a preset rather than a separate engine: it shares the x4plus model of Real-ESRGAN, hosted on fal.ai, with Photo and Basic, face enhancement stays off, and no text-recognition step is involved. For charts and diagrams a better route often exists. If you have the data or the source file, export the chart again at a larger size or as a vector, and every line will be exact rather than predicted.
PNG or JPG for screenshots
PNG is lossless, so every edge in a PNG screenshot is exactly what was on screen. JPG compresses in small blocks and handles sharp, high-contrast edges poorly, which is why JPG screenshots collect halos around letters and speckles across flat areas. Save captures as PNG whenever you have the choice. JPG, PNG and WebP files up to 25 MB are all accepted, so nothing needs converting before upload.
If all you have is a JPG, upload that file as it is. Converting it to PNG does not remove the blocks, it only preserves them, and saving it as a JPG again adds another round of damage. When someone sends you a screenshot through a messaging app, ask for the original sent as a file or document rather than as a photo, since photos are often recompressed on the way while file attachments usually are not.
Screenshot enhancer maths: 2x, 4x and the size ceiling
In the screenshot enhancer, a capture is shrunk in your browser only when it passes the ceiling for its scale: about 6 megapixels for 2x and 2.25 for 4x. A 1920×1080 capture is 2.07 megapixels, under both, so 2x returns exactly 3840×2160 and 4x returns 7680×4320. A 2880×1800 capture (5.2 megapixels) goes in whole at 2x and comes back at 5760×3600, but for 4x it is first reduced to about 1897×1186.
Small captures gain the most. An 800×600 dialog box is processed at full size and becomes 1600×1200 at 2x or 3200×2400 at 4x. Large captures from high-resolution screens work against you at 4x: that reduction can soften small text before any sharpening happens. Crop to the part you need, ideally under 2.25 megapixels, and none of it is thrown away.
2x for high-DPI screens, slides and help docs
High-DPI screens draw each interface pixel with two or more physical pixels, so an image sized for a standard screen looks soft on them. The usual fix is an image twice the size it is displayed at: a screenshot shown 800 pixels wide on a web page should be 1600 pixels wide to look sharp on a 2x display. That makes 2x the natural choice for documentation, help articles and product pages.
Slides follow the same logic. A 960×540 capture at 2x becomes exactly 1920×1080, enough to fill a full-HD slide without softening. 4x is for small crops that must cover a large area, such as a 400×300 detail of a settings panel stretched across half a slide, or for print.
Redact personal and sensitive data before uploading
Screenshots routinely contain more than you meant to share: names, email addresses, phone numbers, profile pictures, home addresses, order and account numbers, API keys, access tokens, QR codes, notification banners and open browser tabs. Crop or cover all of it before you upload, and check the edges of the capture, where status bars and sidebars tend to hold personal details.
Cover sensitive areas with a solid, opaque box rather than a light blur or pixelation. A screenshot enhancer exists to reduce blur, and although it cannot truly recover hidden text, it can make a weakly blurred area look more like writing or fill it with convincing guesses. Redact first and enhance second: guest uploads and results are deleted after 7 days (signed-in results stay in your workspace until you delete them), and anything sensitive should never be in the upload at all.
When re-capturing beats enhancing
If you can still reach the source, a new capture beats any screenshot enhancer. Zoom the page or app to 200% before capturing, or take the screenshot on a high-DPI display, and you get real pixels instead of predicted ones. Export slides and documents straight to image or PDF at high resolution, export charts from their data, and raise a game's render resolution before capturing a scene.
Enhancing is for the cases where going back is not possible: the app has been redesigned, the post or message is gone, the only copy is the compressed one someone sent you, or an old tutorial needs sharper images without redoing every step. There, a cleaner and larger version of what you have is the best result available, as long as you check the text.