How to upscale an image
- Choose an image, drop it on the page, or paste it from your clipboard.
- Pick ×2 or ×4, and the Photo or Illustration model.
- Press Enlarge. The first run downloads the model (about 3–5 MB).
- Drag across the preview to compare with a standard resize, then download the PNG.
Why AI upscaling looks sharper than resizing
A normal resize only spreads existing pixels further apart, so edges go soft and JPEG blocks get bigger. A super-resolution network has learned what real edges, textures and lettering look like, and redraws them at the higher resolution. The result is crisper lines, cleaner skin and foliage, and far fewer compression artefacts.
Private by design
Most online upscalers upload your photo to their servers. This one doesn't: the model runs on your device through WebGL. That makes it safe for personal photos, client work and documents, and it keeps working once the page has loaded.
Frequently asked questions
Is it free to upscale images here?
Yes. You can enlarge images ×2 or ×4 for free, with no account and no watermark.
Are my images uploaded to a server?
No. The neural network is downloaded to your browser and runs on your own device's graphics chip, so your image never leaves your computer or phone.
What's the difference between the Photo and Illustration models?
Photo (Real-ESRGAN general v3) is trained on real-world pictures and cleans up noise and JPEG blocks. Illustration (Real-ESRGAN anime v3) keeps flat colours and crisp lines, so it suits drawings, anime, logos and screenshots of graphics.
Which images benefit most from AI upscaling?
Small, soft or heavily compressed images: old web photos, thumbnails, screenshots and messaging-app pictures. A photo that is already large and sharp changes less, because there is little missing detail to rebuild.
Is there a size limit?
The enlarged result can be up to 16 megapixels (for example 4096 × 4096). Larger inputs are reduced first, and the page tells you when that happens. Large images are processed in tiles, so they take longer but don't run out of memory.
What format is the result?
A lossless PNG. Transparency is kept if your original had it.
Models: Real-ESRGAN realesr-general-x4v3 and realesr-animevideov3 by Xintao Wang et al., BSD-3-Clause licence, running on TensorFlow.js.