Enlarging a small icon or avatar
What limits a tiny avatar is usually the copy you kept, not the tool. Find the largest original before enlarging.
Steps
- Find the largest copy you have
Look for the export from the paint program or the original camera file before reaching for the version saved from a profile page. - Work out the size the destination needs
Multiply the displayed size by two, because most screens draw an image at twice its layout size. - Enlarge by the smallest step that reaches it
Open Pixmend, pick x2 when doubling is enough and x4 when it is not, and choose the illustration model for drawn avatars. - Export as PNG and keep that file
Save the enlarged file yourself, since a platform will re-encode whatever you upload to it.
An avatar looks bad at a larger size for a reason that is decided before any tool runs: the copy you have has usually been through a platform's resizer and its compressor, and that file is not the one you originally made. Finding the largest version you own is worth more than any setting, and sometimes it ends the problem outright. When it does not, open Pixmend and enlarge the best copy you found.
Look for the original first
The picture saved from a profile page is at least two generations down. It was resized on the way in, re-compressed for delivery, and possibly re-compressed again by whatever you used to save it. Enlarging that is enlarging the damage along with the drawing.
Before anything else, check for the export from the paint program, the photo the crop was taken from, or the file you uploaded in the first place. A 512 px original sitting in a folder often already meets the requirement, and no processing is needed at all. Enlargement is what you do when the original is genuinely gone.
Work out the size you need, then double it
Most screens draw an image at twice its layout size, so an avatar shown in a 200 px circle wants a 400 px file behind it. That single factor accounts for most of the "it looks soft on my laptop but fine on my phone" reports.
The arithmetic is the whole decision:
| Where it appears | Displayed at | File you want |
|---|---|---|
| Comment or chat avatar | 40 px | 80 px |
| Profile header | 200 px | 400 px |
| Stream overlay or card | 400 px | 800 px |
| App or store icon | 512 px | 1024 px |
Divide the file you want by the file you have. If the ratio is two or less, x2 covers it; up to four, x4 does. A 256 px icon becomes 1024 px at x4, which is enough for the largest row in that table. Beyond four, enlargement is not the answer — the detail to fill that gap is not in the file, and anything that appears there was invented rather than recovered.
Pick the smallest step that clears the requirement rather than the largest one available. Note that x2 and x4 take the same time, because x2 is the x4 result scaled back down, so choosing x2 is a decision about the picture and never about waiting less.
Drawn avatars and photographed avatars need different models
They fail differently, so they are handled differently. A drawn avatar is flat colour bounded by lines, and the model tuned for illustration keeps those edges crisp instead of treating them as noise to smooth. A cropped photograph has grain and soft gradients, and the photo model is the one that respects them. Using the photo model on line art is the usual cause of an avatar coming back slightly melted.
Noise removal follows from where the file has been rather than from what it shows. A PNG straight out of a paint program has no compression damage to remove, so leave the setting alone. An icon that has been through a chat app or a profile page is a JPEG several times over, and a moderate setting cleans the blocks and the shimmer around the linework as it is enlarged.
If the icon has transparency, keep it in PNG from start to finish. Transparent pixels still carry colour underneath, and a pipeline that discards it leaves a dark fringe once the image is enlarged.
What this cannot do
It repairs what is in the file; it does not draw what is missing. A face at 64 px across has lost the features, not merely the sharpness of them, and no honest process returns them. Enlarged, it is the same missing detail at a larger size — which is exactly the point at which generative tools hand back a stranger's face instead.
There is also a limit at the other end of the trip. Whatever you upload, the platform will resize and re-encode it to its own specification, so the clean file you made is not the file that gets served. Keep your own copy: that is the one worth having for a video thumbnail, a printed sticker, a slide, or the next site that asks for a bigger picture than the last one.
Practical notes
Icons are small enough that none of the limits here matter much. The 12.6MP cap on input is orders of magnitude above any avatar, and speed is a non-issue: a full 1024×1024 photograph takes 7.1 seconds, and an illustration of that size 3.9 seconds, so a 256 px icon finishes well inside that.
The first run downloads about 11.6MB of models and runtime, which is then kept on your device. The image itself never leaves the browser — it is processed on your own GPU, and you can confirm that by watching the Network tab in DevTools stay silent while the work happens. There is no account, no install, and no limit on how many times you do it.