Image upscale · sharpening
Zoom the small image 2x with AI in the browser and save it as a PNG. A 3·4x output uses additional interpolation, and the image will not be sent outside the device.
Photo to be enlarged
Preparing Public example.
Original and enlarged result
Can immediately run AI zoom in with Public example.
Move the slider to compare before and after expansion.
Share result
Share the tool link and result text together. No URL is included in the result. Check the below content and then share it.
How to zoom a photo with AI
When the page is opened, 160 × 120px example results will be displayed. Click ‘Zoom in’ to generate an AI 2x 320 × 240px PNG. Choose 3x 4 to further expand the AI 2x result using Lanczos interpolation.
Run Swin2SR lightweight x2 model in browser’s WebAssembly executor to create a 2x image. Choose 3x or 4x to further expand AI 2x result using Lanczos interpolation. Neither method uploads the file to a server.
The model estimates new pixels based on pixel patterns in small images. It can create detailed descriptions not present in the original, but does not restore blur, compression noise, or text. The original is limited to a long side of 512px and 260,000 pixels or less. Large images should be processed first. Image size adjustment toolCan be scaled down here.
Frequently Asked Questions
Does AI restore the details of the original photo?
The Swin2SR lightweight model estimates a 2x image based on input pixels. It does not guarantee restoration of original details or the removal of patterns or shapes. For 3x or 4x, Lanczos interpolation is added without AI repetition.
What size of photos can be processed?
JPG·PNG·WEBP supports static images up to 20MB, 512px per long side, and 260,000 pixels or less. Large photos are not processed directly due to model memory limits. You can resize images using the image size tool first
Does the photo file get sent to the server?
No. Image decoding, AI inference, and PNG generation are processed in the browser and device. On first execution, you can download the model and WebAssembly executable from this site, but no image file uploads are sent.
Calculation criteria and reference materials
- CAIDAS — Swin2SR lightweight x2 image super-resolution model
An 2x image model based on Apache-2.0, but it does not restore loss or blur, nor guarantee individual photo quality.
- Hugging Face Transformers.js — Image-to-Image pipeline
API that runs the image-to-image model in the browser and returns RawImage results.
- MDN — OffscreenCanvas.convertToBlob()
Encode canvas data inside the walker as a PNG file
Verification of calculation criteria: · Calculation·verification principles · Report errors