The Upscale Button in Your Design App vs Upscayl vs a Browser Upscaler: Which to Use?

There are now three common ways to enlarge an image: the upscale button built into the design or photo app you already use, a desktop app such as Upscayl, or a browser-based upscaler like ours. They're not competing for exactly the same job. This post explains what each one is good at, what it costs you in money, time and privacy, and how to pick, without pretending one of them wins every time.

Option 1: the upscale button in an app you already use

Many design editors, photo editors and mobile photo apps have added a one-click upscale or "enhance" feature. The appeal is obvious: the image is already open, the button is right there, and the result lands back in your design.

What it's good at: convenience. If you're building a slide or a social post inside an editor and one picture is a bit small, upscaling it in place saves exporting, processing and re-importing.

What to check:

• Whether it's included in your plan. Upscaling is often a premium feature or limited by credits.
• The output size. Some built-in upscalers have a fixed factor or a maximum size. Look at the exported dimensions, not just the preview.
• Where the processing happens. In most online editors your image is processed on their servers. That's usually fine, but worth knowing for private photos.
• What happens on export. An editor may re-compress the image when you export the finished design, so the extra pixels can be partly undone by a low export quality setting.

Option 2: a desktop app like Upscayl

Upscayl is a free, open-source upscaler for Windows, macOS and Linux, released under the AGPL-3.0 licence. It runs Real-ESRGAN models on your own graphics card through the Vulkan API and includes a batch mode for folders of images.

What it's good at: volume and model size. Because it runs natively on your GPU, it can use larger models than a browser can, and it doesn't care how many images you throw at it. Nothing leaves your computer.

What to check: its documentation states that it needs a Vulkan-compatible graphics card, and that it generally won't run on most integrated graphics or on the CPU. It also needs installing and updating, which is a small cost for a regular user and a real hurdle on a locked-down work laptop. Like every AI upscaler, its own FAQ notes that it guesses missing detail and can't fix a photo that was out of focus to begin with.

Option 3: a browser upscaler

Browser upscalers come in two kinds: those that upload your image to a server, and those that run the model in your browser. Ours is the second kind. It downloads a compact super-resolution model (ESRGAN-slim, through the UpscalerJS library) once, then processes images on your computer's graphics card through WebGL.

What it's good at: no installation, no account, no upload, no limits. It works on any computer with a modern browser, including ones where you can't install software. On our test computer a 480-pixel image took 3 to 6 seconds at 4x once the model was loaded, and a 1.7-megapixel photo about 20 seconds at 2x.

What to check: browsers can't run models as large as a desktop app with a strong GPU, so on the hardest cases a desktop app may do better. On phones and tablets, our tool uses a classic (non-AI) engine, because the model needs more graphics memory than mobile browsers allow. And the size limits: the AI handles source images up to about 10 megapixels and the tool is made for results up to 4K.

BeforeAfter

Side by side

What actually decides the result

Comparisons between upscalers often focus on the model. In our testing, the model mattered less than three other things.

The input. The same photo enlarged from the camera original and from a messaging-app copy gave very different results, whatever tool did the enlarging. Compression blocks get enlarged along with everything else. Find the best original first.

The factor. Asking for 4x when 2x reaches the size you need gives you a slower process and a bigger file, with nothing extra to see. Divide the width you need by the width you have and round up.

The output format. A photo enlarged 2x was 6.85 MB as PNG and 337 KB as WebP in our tests, and looked the same. If a tool only exports PNG, expect large files; if you need a smaller one, our image compressor can aim at a size in KB without changing the dimensions.

How to choose

• You're inside a design and one image is a little small: use the app's button if it's included in your plan, and check the export size and quality.
• You upscale lots of images regularly and have a suitable graphics card: install a desktop app.
• You need a few images enlarged now, privately, on whatever computer you're at: a browser upscaler that runs locally.
• You're on a phone with a tiny old photo: do it on a computer if you can; phone browsers can't run the larger models.

Whatever you use, check the result at 100% on the parts that matter (faces, text, product edges) before you print or publish it. AI upscalers rebuild edges convincingly and invent fine texture; that's fine for a wall print and not something to rely on for reading a number plate.

What happens to your file in each case

Quality comparisons get all the attention, but for many people the deciding question is simpler: where does my photo go?

• Server-side tools, including most built-in buttons in online editors, receive your original file. That includes everything hidden in it: phone photos usually carry the GPS location where they were taken, the date and time, and the device model. Reputable services handle this responsibly, but the file has left your computer.
• Desktop apps process the file where it is. Nothing is sent anywhere unless you choose to share the result.
• Browser tools that run locally read the file in the browser and never upload it. In our checks, the files our upscaler, compressor and cropper produce also carry no camera data at all, because each one saves a completely new image. When we upscaled an iPhone photo that contained GPS coordinates, the result had no location, no camera model and no date left in it.

If you're sharing an original photo rather than a processed copy, remove its location first; the metadata remover does that without re-encoding the picture, and the guide on how to remove location from a photo shows what a location tag reveals.

Cost over time

For a single image, cost barely matters. For regular use it adds up differently for each option. Credit-based upscaling is cheap per image and adds up with volume. A desktop app is free but costs setup time and, if your computer lacks a suitable graphics card, possibly hardware. A browser tool that runs locally costs nothing per image and needs no setup, but uses your computer's time: on our test machine, 3 to 20 seconds per image depending on its size.

A sensible approach for most people is to start with the free option that needs no account, and only move to a paid feature or a dedicated app when you hit a case it can't handle well: very large batches, tiny faces, or images that need more than a 4x enlargement in one go.

One more practical difference: what happens when something goes wrong. A built-in feature fails quietly or asks for credits; a desktop app may not start on unsupported graphics hardware; a browser tool may fall back to a simpler engine. Ours tells you after each job whether the AI or the classic engine ran, so you know what you got.

A quick way to compare them on your own photos

If you have access to more than one option, the fairest comparison takes five minutes. Take a sharp photo of the kind you usually enlarge, shrink a copy to a quarter of its width, then enlarge that copy 4x in each tool. Compare each result with the original at 100%, looking at the same spots each time: a face, a fine edge, any text. The original is the right answer, so the closest result wins. Keep the export format the same for every tool and leave extra sharpening off, so you're comparing the models rather than their settings.

Combining tools

These options aren't exclusive. A common workflow: enlarge in a dedicated upscaler, then place the result in your design app, which keeps the design app's export from working on a too-small source. If the image also needs a new background, upscale first and then remove the background. When we tested both orders on a small product photo, upscaling first gave edges about 20% crisper; the full test is in our post on whether to remove the background or upscale first.

For specific targets, the guides are more useful than any comparison: upscaling an image to 4K for screens and wallpapers, increasing image resolution for print, and white background product photos for shops.

Frequently asked questions

Do built-in upscale buttons upload my photo?

In most online editors the image is processed on the company's servers. Desktop apps and browser tools that run locally, like ours, keep the file on your device.

What hardware does Upscayl need?

Its documentation says it needs a Vulkan-compatible graphics card and generally won't run on most integrated graphics or on the CPU.

Should I upscale before placing an image in a design?

Yes, if the image is too small: enlarge it first in an upscaler, then place it, so the design is exported from a large enough source.