AI Image Upscaler

PixelShift uses the ESRGAN Thick neural network to reconstruct detail at 2× resolution. It is not a simple bicubic stretch—it runs inference locally in your browser via TensorFlow.js and WebGL.

No. Image pixels and the AI model are processed on your device. PixelShift does not send your photos to a cloud upscaling API.

Neural upscaling is compute-intensive. PixelShift processes one image at a time in a Web Worker and shows per-tile progress so the page stays responsive.

The ESRGAN model (~28 MB) downloads once to this browser on first use. Your selected images stay in the queue with a progress bar, and upscaling starts automatically when the download finishes. Later visits reuse the cached files.

The longest side of your source image can be up to 2048 pixels for 2× upscaling. Images are processed in small GPU tiles. Larger images are rejected before processing to avoid running out of GPU memory.