Frequently Asked Questions

On a desktop or laptop it runs ESRGAN-slim, a super-resolution neural network, on your graphics card through WebGL. The model predicts the detail that a plain resize would blur, so edges, hair and text come out sharper. In our tests it produced about eleven times more edge detail than an ordinary resize of the same photo.

Divide the width you need by the width you have and round up. Full HD (1920 × 1080) to 4K is exactly 2x. Choosing 4x when 2x reaches your target only gives you a slower job and a bigger file. Our guide on how to increase image resolution lists the pixels each print size needs.

Phones and tablets always use the classic engine, because mobile browsers don't give web pages enough graphics memory for the AI model. On a computer, the classic engine is used for photos above about 10 megapixels, which rarely need upscaling anyway.

Up to 4K: a Full HD image reaches 4K at 2x, and smaller images can use 4x. If your computer can't hold the result, the tool reduces the size automatically and tells you.

WebP for the smallest file, JPG if you're not sure where the image is going, PNG for graphics or further editing. In our tests a 2x photo was 6.85 MB as PNG and 337 KB as WebP, with no visible difference.

Yes. The Bulk tab takes up to 20 images (each up to 100 MB) with the same settings and downloads them as one ZIP file.

Yes. A PNG or WebP with transparency stays transparent when you download PNG or WebP. If you choose JPG, which can't store transparency, the transparent area becomes white.

It helps with photos that are small or soft from being shrunk. It can't undo camera shake or missed focus, because that detail was never recorded.

Free, Square 1:1, Landscape 1.91:1, Portrait 4:5, Widescreen 16:9, Photo 3:2, Passport 35×45, Photo 3:4 and Banner 2:1. You can also type any custom ratio, such as 50 : 70 for Canadian passport photos or 9 : 16 for Stories.

Choose Passport 35×45 for the 3.5 × 4.5 cm size used by most countries, or a custom ratio for countries with another size. The passport photo size guide lists sizes and head heights by country.

Type the width and height under Exact size, for example 1080 × 1350, and press Set. The selection locks to that shape and the download is exactly that many pixels. The Selection readout below the photo shows how many pixels of the original you are using.

Cropping keeps the original pixels inside the frame at full resolution. The file is saved again in the format you choose, so pick PNG if you want no compression at all.

JPG can't store transparency, so the transparent area is filled with white instead of black.

No. The exported file contains no camera metadata, including GPS. We checked every file from our tests.

Yes. Choose Circle: the selection becomes a circle and the download is a square image with the corners cut away, transparent as PNG or WebP and white as JPG. For profile pictures on social networks a square crop is enough, because they display it as a circle anyway.

Quality lets you choose a quality level and shows an estimate of the final size. Target Size takes the size you need, for example 100 KB, and finds the best quality that fits, reducing the dimensions only if quality alone isn't enough.

The compressor counts 1 KB as 1024 bytes, while some upload forms count 1000. A file of 100 KB in our terms is 102,400 bytes, which a strict form would reject. Typing 97 keeps you safely under either way. See the guide to compressing an image to 100 KB.

Yes. Type a width in pixels and the image is resized before compression, which is often the best way to reach a small size with good quality.

JPG is accepted everywhere. WebP is smaller at the same quality and keeps transparency. PNG is lossless and best for screenshots and graphics.

It depends on the photo. In our tests an 8-megapixel photo couldn't reach 100 KB by quality alone, so the tool also reduced its dimensions. Very small targets like 20 KB suit ID photos and signatures more than detailed photos.

Yes. Compressed files come out with no camera data or GPS location, because the image is saved as a new file.

It runs an image segmentation model with ONNX Runtime in your browser: BiRefNet on the graphics card through WebGPU where available, otherwise the lighter ISNet. The model (about 115 MB on a computer, 88 MB on a phone) is downloaded once and cached, so later visits start quickly.

Transparent (downloaded as PNG), White, or any Color you pick, downloaded as PNG or JPG. The result always has the same dimensions as your original.

People, products and animals that stand out from their background. In our tests a mug, a dog, a portrait and a person in a cap all came out cleanly. When the subject is low-contrast against something more eye-catching, crop closer to the subject first.

Often, but check the result. In our test it kept the white centre inside a ring-shaped logo and dropped a separate underline from a signature. For flat graphics on pure white, a one-colour tool works better; the guide on how to make a background transparent explains the options.

About 1.5 to 2.5 seconds for a typical photo on a computer with WebGPU once the model is loaded; the first photo after opening the page takes a few seconds more. Phones and browsers without WebGPU use a lighter model, about ten seconds on the processor alone. The first run includes the model download.

PNG stores every pixel losslessly. Save it as WebP in the image compressor to keep transparency in a much smaller file: our 2.5 MB cut-out became 220 KB as WebP.

EXIF (camera, settings, dates), GPS location, XMP (editing history and more) and IPTC (captions, copyright). The report shows what each photo contains before you remove it.

No. It removes the metadata blocks and leaves the image data byte for byte the same. We compared every pixel before and after in our tests.

Keep photo rotation preserves the single tag that keeps phone photos upright. Keep color profile preserves the colour information so colours look the same. Both are on by default.

Yes. Add up to 50 photos at a time; the cleaned files download together as a ZIP.

JPG, PNG and WebP.

It removes the hidden data, not what's visible in the picture. House numbers, street signs and reflections still show a location; crop them out. The guide on how to remove location from a photo covers both.

Yes. It detects and removes Content Credentials (C2PA), the IPTC label that marks an image as AI-generated, and the prompt and settings that tools such as Stable Diffusion save in PNG files. It can't remove invisible watermarks stored in the pixels themselves, and removing a label doesn't change the rules about disclosing AI images where you post them.

Website URL or text, email, phone number, SMS and Wi-Fi network login.

No. The information is stored in the code itself, so it works as long as the content (for example a web address or Wi-Fi password) stays the same. There's no tracking and no redirect service.

PNG (256 to 2048 pixels) for screens and documents, SVG for printing at any size.

Yes. Keep a dark code on a light background with strong contrast. The generator warns you if the colours are inverted or the contrast is too low to scan reliably.

Yes. Special characters such as ; , : " and \ are escaped automatically, and hidden networks are supported. The WiFi QR code guide explains security types and printing.

Medium suits most codes. Choose High if the code will be printed small, on a curved surface or with a logo over part of it.

No. All six tools process images in your browser. The AI tools download their engine and model once; your photos never leave your device.

Yes. No account, no credits, no watermark and no daily limit. The site is supported by advertising.

Yes, in any modern mobile browser. The upscaler uses a lighter engine on phones; for small or old photos, a computer gives better results.

English, Spanish, French, German, Italian, Portuguese, Japanese and Korean. Every tool, guide and article is translated.