Upscale an image with AI

Make a small picture bigger and sharper with a neural network that runs on this device.

Makes a small or blurry picture 2, 3 or 4 times bigger with an AI upscaler, for sharper edges and fewer JPEG blocks. It is free, needs no account and adds no watermark, and your files are not uploaded: everything runs in your browser on your device. The first run downloads the upscaling model (about 4.9 MB) and the WebAssembly engine that runs it to your browser, and after that the tool works offline.

Drop pictures to upscaleor click to choose files. Batches welcome.

    Pictures are held under 16 megapixels, the most every browser can save.

    Working…

    Done

    Your files are processed on this device and never sent to a server. Reload the page to clear everything from memory.

    How it works

    1. Drop one or more pictures (JPG, PNG, WebP, HEIC, AVIF or JPEG XL).
    2. Pick 2×, 3× or 4×, and keep the format or choose PNG, JPG or WebP.
    3. Run it. The upscaling model downloads once (4.9 MB), then each picture is enlarged in your tab.

    In your browser, not on a server

    Your browser decodes the picture. Real-ESRGAN general x4v3 (BSD-3-Clause, Xintao Wang and others), a compact neural network, runs in a web worker through onnxruntime-web (MIT) compiled to WebAssembly. It enlarges the picture 4× in overlapping 128-pixel tiles whose seams are cut away; for 2× and 3× the picture is first shrunk so that 4× lands on the size you asked for. Transparency is enlarged separately and put back. The result is saved by your browser’s own image encoder.

    Compared with upload-based tools: Online AI upscalers upload your photo to a GPU server, limit free use to a few images a day, add watermarks or ask for an account, and keep the upload. Here the network runs on your device and the picture never leaves it.

    Verify it yourself. Give the page a short while after your first run to save itself for offline use, then turn off Wi-Fi or switch to airplane mode and run the tool again. Or open your browser's developer tools (F12, or Cmd-Option-I on a Mac), pick the Network tab, and watch it while you convert. The panel above does the same count for you.

    Questions

    Are my pictures uploaded?

    No. The upscaler is a neural network that runs as WebAssembly inside this page, in a background thread. Your pictures are decoded, enlarged and saved in your tab. The model and the WebAssembly runtime that runs it are downloaded once, from this site.

    How is this different from resizing?

    Resizing spreads the same pixels further apart, so a small picture gets soft and blocky. The upscaler is Real-ESRGAN, a network trained on pairs of sharp and degraded photos, so it draws crisp edges and removes JPEG blocks and blur while it enlarges.

    Can it recover a face or text that is not in the picture?

    No. It sharpens what is there and fills in plausible fine texture, but it cannot restore detail that was never captured. Tiny faces and small print come out cleaner, not readable if they were not readable before.

    How big can the result be?

    Up to 16 megapixels, about 4900 × 3300, because that is the most every browser, including Safari on iPhone, can hold as one image. A larger request is held to that size and the result says so.

    How long does it take?

    Time grows with the size of the result. A 4× enlargement to about 2000 × 1500 takes around half a minute on a recent laptop and longer on a phone. The page stays usable while it works, and several pictures run one after another.

    More image tools

    Popular conversions

    Guides