Tools › On-device AI
Upscale an image with AI
Enlarge photos and screenshots 2× or 4× with a super-resolution model that runs in your browser. Large images are processed tile by tile.
The model downloads once to your device; your photo never leaves it. Nothing starts until you choose an image.
How it works
- The model downloads once to your device; your photo never leaves it.
- The image is split into tiles with overlapping borders, so big pictures fit in GPU memory and show no seams.
- 4× uses a model trained on real-world photos with noise and JPEG artifacts; 2× uses a lighter model for clean images and screenshots.
- Runs on WebGPU when your browser has it and falls back to WebAssembly on the processor.
Questions
Is my image uploaded?
No. The model downloads once to your device; your photo never leaves it. Only the AI engine (from this site) and the model weights (from Hugging Face) are downloaded; your picture stays in this tab.
How long does it take?
With WebGPU, seconds for a typical photo. Without it, the processor does the work and a 1-megapixel photo at 4× can take several minutes. You can cancel at any time.
What is the largest image?
The result can be up to 64 megapixels and 16,384 pixels on the long side, the limit browsers allow for images.
Which model is used?
Swin2SR (Apache 2.0 licence). You may use the results commercially.