Skip to content
FitMyImage
More tools
English

Upscale an image with AI

Enlarge photos, screenshots and artwork two or four times with an AI model that redraws the detail instead of just stretching the pixels. It runs on your device; nothing is uploaded.

Enlarge
Noise reduction

The AI model (about 5 MB) downloads once and runs on this device's graphics chip. Photos are never uploaded. A 1 MP photo takes about 10 to 30 seconds.

or drop images hereJPG, PNG, WebP, HEIC, GIF, TIFF, PSD, RAW and SVG. Nothing is uploaded. You can also paste a screenshot.

How it works

  1. Choose 2× or 4×. 2× suits most photos; 4× is for small images such as thumbnails, icons and old web pictures.
  2. Set the noise reduction. Medium works for most photos; Low keeps film grain and texture, High cleans up noisy or heavily compressed images.
  3. Drop the images. The first one also downloads the model, about 5 MB, once.
  4. Use Compare on a finished row to check the detail, then download one or all as a ZIP.

What the AI does

Ordinary resizing spreads each pixel over a larger area, so edges go soft and JPEG blocks get bigger. The upscaler here is Real-ESRGAN, a neural network trained on millions of image pairs to predict what the sharper version looks like: it rebuilds edges, lines and textures and removes compression noise as it enlarges. It cannot recover what was never in the picture, such as a readable number plate from a few pixels, but it makes small images look clean at a larger size.

Size limits

Browsers refuse canvases above about 16 million pixels, so a 4× result is limited to roughly 4000 × 4000 pixels. A larger input is made smaller first and the row says so. For print, upscale 2× and set the DPI with the resize tool.

Questions

Is the upscaling really done in my browser?

Yes. The model's weights are downloaded from this site and run with TensorFlow.js on your graphics chip through WebGL. The image is read, enlarged and saved inside the tab; there is no server that receives it.

How long does it take?

About 10 to 30 seconds for a 1 megapixel photo on a recent laptop or phone, longer at 4× or on older devices. The row shows the progress. Without GPU access the browser enlarges the image without AI and says so.

Should I choose 2× or 4×?

Use 2× when the image is already a reasonable size and you need more pixels for printing or a bigger screen. Use 4× for small pictures: icons, thumbnails, screenshots from old games, scanned stamps. Both use the same network; 2× is the 4× result scaled down, which keeps the extra detail.

What does noise reduction change?

It blends two versions of the model. High smooths away grain, sensor noise and JPEG artefacts, which suits phone photos taken in low light and images saved many times. Low keeps fine texture such as skin, fabric and film grain, at the cost of keeping some noise.

Can it make a blurry photo sharp?

It sharpens edges and removes compression blur, and small or slightly soft images improve a lot. Motion blur and missed focus are a different problem: the detail is gone from the photo, and the model can only guess at it.

Related tools