AI image upscaler
Enlarge photos 2× or 4× with AI that runs on your own device.
next step Image compressor
report a problemHow it works
- 01Pick 2× (quick, for clean images) or 4× (for photos), then download the model once.
- 02Drop in a JPG, PNG, WebP or HEIC. The image is upscaled tile by tile, and you can watch it sharpen.
- 03Zoom in to compare before and after, then download it as JPG, PNG or WebP.
Questions
›Which AI model does this use?
Swin2SR by Marcos V. Conde et al., released under the Apache 2.0 license. 2× uses the lightweight model (about 8.1 MB); 4× uses the real-world model trained on degraded photos (about 54 MB on a graphics card, 19 MB on the processor). We use the ONNX conversions published on Hugging Face and run them with Transformers.js.
›How long does it take?
With WebGPU (recent Chrome or Edge on a computer), a typical photo takes seconds to a minute. Without it, the model runs on the processor: 2× takes a few minutes for a 1000 px image, and 4× can take ten minutes or more. The page shows an estimate before you start and the time left as it works. The first visit also downloads the model once; after that it loads from your browser.
›Why is there a size limit?
The limit is on the result. 4× turns a 1500 px image into 6000 px (up to 36 megapixels), and a browser tab runs short of memory beyond that. We work on images up to 1500 px for 4× and 3000 px for 2× (1024 and 2048 px on phones). Bigger images can be shrunk first, or cropped to the part you need.
›Will it really add detail?
It's AI, so it predicts detail rather than recovering it. Edges get crisper and blur, noise and JPEG blocks are cleaned up, but faces, small text and fine patterns can come out slightly different from the real thing. Zoom in and check before you print or publish.
›Are my files uploaded anywhere?
No. Your files are processed on your own device by your browser. They never touch our servers, so there are no upload waits, no size caps from us, and nothing to delete afterwards.