Skip to content
Kordu Tools Kordu Tools

AI Image Upscaler

AI Runs in browser

Upscale images 2x or 4x with AI — sharper than bicubic, runs in your browser, no upload needed.

Last updated 01 Apr 2026

Upscale photos using Swin2SR — a Swin Transformer super-resolution model that reconstructs realistic texture and detail at 2x or 4x scale. Far sharper than standard bicubic or Lanczos resampling. Runs entirely in your browser using ONNX Runtime Web — the ~20MB model downloads once and caches for instant reuse. Your images are never uploaded to any server.

Scale Factor

2x doubles each dimension. 4x runs two passes and quadruples dimensions — significantly slower.

~19.1 MB download

Click to upload or drag and drop

JPG, JPEG, PNG, WEBP up to 10MB
Loading rating…

How to use

  1. 1

    Choose a scale factor

    Select 2x to double both dimensions or 4x to quadruple them. 4x runs two successive AI passes and takes significantly longer — start with 2x to preview quality.

  2. 2

    Upload your image

    Drag and drop or click to upload a JPG, PNG, or WebP image up to 10MB. Images under 1024×1024 process fastest.

  3. 3

    Wait for AI processing

    Swin2SR processes your image through a deep neural network. On first use, the ~20MB model downloads and caches. Subsequent upscales load instantly.

  4. 4

    Compare before and after

    Drag the before/after comparison slider to examine the difference in detail and sharpness between the original and the upscaled result.

  5. 5

    Download the upscaled image

    Click Download to save the upscaled result as PNG.

Frequently asked questions

Is my image uploaded to a server?
No. All processing runs entirely in your browser using ONNX Runtime Web and WebAssembly. Your images never leave your device — processing is completely private.
What model is used for upscaling?
Swin2SR — a Swin Transformer v2 model trained for classical super-resolution. It reconstructs realistic textures and sharp edges that bicubic interpolation cannot recover by analysing learned patterns from millions of training images.
How large is the AI model download?
Approximately 20MB. It downloads once on first use and is cached in your browser — future upscales are instant with no re-download.
What is the difference between 2x and 4x upscaling?
2x doubles width and height (4× total pixels). 4x quadruples each dimension by running two 2x passes (16× total pixels). 4x takes significantly longer and produces a much larger output file.
What image sizes work best?
Input images under 1024×1024 process fastest. Larger images will still upscale correctly but may be slow on lower-end devices or mobile browsers.
How does this compare to Let's Enhance or Topaz Gigapixel?
Those services upload your images to their cloud servers for processing. This tool runs entirely in your browser — your images stay private. Quality is comparable for most photo and illustration types.
Why does upscaling take so long?
Super-resolution AI analyses every tile of the image through a deep transformer network with hundreds of millions of parameters. It's fundamentally different from a simple resize — each output pixel is computed from learned patterns. Processing time depends heavily on device CPU performance.
Can I upscale illustrations, screenshots, and diagrams?
Yes. Swin2SR was trained on diverse image types and handles natural photos, digital art, and compressed images well. Results vary by content — photos and illustrations generally upscale better than text-heavy screenshots.
Does upscaling work offline?
Yes. Once the model has been downloaded and cached by your browser it works completely offline without any internet connection.

AI image upscaling uses deep learning to intelligently reconstruct detail when enlarging

a photo — producing dramatically sharper results than traditional bicubic or Lanczos

resampling. This tool runs Swin2SR, a Swin Transformer v2 super-resolution model trained

specifically to recover fine textures, edges, and colour gradients from low-resolution inputs.

2x upscaling doubles both width and height (4× total pixels) — ideal for printing small

photos at larger sizes, sharpening compressed web images, enlarging avatars and icons without

visible pixelation, and enhancing product photos. 4x upscaling quadruples each dimension

by running two successive 2x passes, suited for heavily downscaled originals, archival

restoration, and maximum output resolution requirements.

Unlike Let's Enhance, Topaz Gigapixel, or AI.Image.Enlarger — all of which upload your

images to their servers — everything here runs locally in your browser using ONNX Runtime

Web (WebAssembly). Your images never leave your device. The Swin2SR model (~20MB) downloads

once and is cached by the browser, so subsequent upscales load instantly.

Works on photos, digital illustrations, compressed JPEG/WebP images, and screenshots.

Input images under 1024×1024 process fastest. Larger inputs work but may be slow on

lower-end devices.

Related tools

Image Resizer

Resize images to exact pixel dimensions or percentage — aspect ratio lock, social media presets, no upload needed.

Image Compressor

Compress PNG, JPG, WebP, AVIF, GIF, BMP, ICO and more — reduce file size without losing visual clarity.

Background Remover

Remove image backgrounds instantly using AI — transparent PNG output, no account needed, runs entirely in your browser.

Depth Map Generator

Generate depth maps from any photo using AI. Visualize 3D structure with colorized depth estimation — runs in your browser.

Image Filters

Apply photo filters and adjustments — brightness, contrast, saturation, grayscale, sepia, vintage, and more.

Learn more

DPI vs PPI: Image Resolution Explained for Print and Screen

DPI vs PPI explained with real pixel math. Learn why 300 DPI matters for print, what PPI means for screens, and how to export images correctly.