AI Image Upscaler

Make a small or soft picture bigger and sharper. The AI enlarges it 2× or 4× and draws in crisp edges and texture instead of the blur you get from a normal resize, and it runs in your browser, so your images are never uploaded.

The AI runs in your browser. Images are never uploaded. Free, no sign-up, no watermark.

Drop images here

JPG, PNG, WebP, HEIC, AVIF, BMP, TIFF · up to 10 images

Options

Upscale

Up to 2048 px on the longest side for ×2; bigger images are scaled down first.

Noise reduction

Keeps fine texture such as skin, hair and fabric.

Save as

JPG is much smaller. Transparent areas become white.

The AI model (about 5 MB) downloads once, then your browser keeps it for next time.

AI upscaling draws plausible detail. It cannot bring back text or faces that are not in the picture, and it will not fix heavy blur.

Photos still called IMG_4821.jpg?

File Renamer AI names photos by what's in them, in bulk. 25 free renames.

How to use it

  1. Drop up to 10 images onto the page or click Choose images. JPG, PNG, WebP, HEIC (iPhone), AVIF, BMP and TIFF all work.
  2. Pick 2× or 4× and the noise reduction: Normal keeps fine texture, Strong cleans up grainy or heavily compressed JPGs.
  3. Click Upscale. The first time, the AI model (about 5 MB) downloads and is then kept by your browser. The progress bar counts the parts of the image as they are done, and you can stop at any time.
  4. Drag the slider across the preview to compare the original with the result, and zoom to 2× or 4× to check the detail. Download a JPG or PNG; several images download together as one ZIP file.

About this tool

Upscaling helps when a picture is too small for what you need: an old photo or a picture saved from a chat app that looks soft when printed, a logo or product shot that needs to fill a bigger slot on a website, a screenshot for a slide, or artwork for a poster. A normal resize only spreads the existing pixels out, so edges get blurry and blocky. An AI upscaler has learned what sharp photos look like and fills in plausible fine detail.

This tool uses Real-ESRGAN, an open-source model (BSD licence) trained to undo the blur, noise and JPEG blocks of real-world photos. It runs inside your browser with ONNX Runtime, on your graphics card where the browser supports WebGPU and otherwise on the processor. The image is processed in overlapping tiles that are blended together, so even a large picture fits in memory without visible seams. On the processor a small photo takes seconds and a large one can take a few minutes.

Be realistic about what it can do. The extra detail is invented, not recovered: it cannot read a blurred number plate, bring back text that is not legible in the original, or restore a face that is only a few pixels wide, and faces can come out smooth or slightly plastic. Heavy motion blur or an out-of-focus shot stays blurry. It works best on photos, screenshots, illustrations and logos that are a little small or soft.

Good to know

  • Maximum input: 1024 pixels on the longest side for 4× and 2048 pixels for 2×, so the result is at most 4096 pixels. Larger images are scaled down to that size first, and you are told when that happens.
  • JPG downloads are much smaller than PNG. PNG is lossless and keeps transparency; on JPG, transparent areas become white.
  • Use Strong noise reduction for grainy low-light photos and heavily compressed images from chat apps; keep Normal for clean photos, where Strong can make skin and grass look waxy.
  • Nothing is uploaded: the model is downloaded once (about 5 MB) and then runs on your device.

Other ways to do it

Photoshop or Lightroom (Super Resolution)

  1. In Lightroom or Lightroom Classic, right-click the photo and choose Enhance, or open a JPG or raw file in Photoshop’s Camera Raw and right-click its thumbnail.
  2. Tick Super Resolution and click Enhance. It doubles the width and height (4× the pixels) using Adobe’s own AI model.
  3. The result is saved as a new DNG file next to the original; export it as JPG or PNG.

GIMP (free, Windows, Mac and Linux)

  1. Open the image and choose Image > Scale Image.
  2. Enter the new width (keep the chain linked so the height follows) and set Interpolation to NoHalo or LoHalo.
  3. Click Scale, then File > Export As. This is a classic resize: smoother than the default, but it does not add detail like an AI upscaler.

Mac (Preview)

  1. Open the image in Preview and choose Tools > Adjust Size.
  2. Make sure Resample image is ticked, then enter a larger width or a percentage.
  3. Click OK and save. Preview only enlarges the pixels, so expect a softer result than an AI upscaler.

Private by design: nothing is uploaded

Most AI upscalers upload your photo to a server, run the model there and send back the result, often with a watermark, a size limit or a credit system. Here the AI model runs inside your browser: the only downloads are the model and the engine (public files, fetched once and cached), and your images never leave your device.

Step-by-step guides

See all free image tools

Frequently asked questions

Is this image upscaler really free?

Yes. There is no sign-up, no watermark, no credit system and no daily limit, and you download the full-size result. You can upscale up to 10 images at a time and start again as often as you like.

Are my images uploaded to a server?

No. The AI model is downloaded to your browser and runs on your own device, so your images and the results never leave it. The only downloads are public files: the model (about 5 MB) from Hugging Face and the ONNX Runtime engine from a CDN.

Can it fix a blurry photo or make an unreadable face or text clear?

Only to a point. It sharpens pictures that are slightly soft or small, removes noise and JPEG blocks, and draws in believable texture. It cannot recover detail that is not in the image: a heavily blurred or out-of-focus photo stays blurry, unreadable text stays unreadable, and tiny faces may come out smooth or look artificial. Do not use it to identify people or read plates and documents.

Should I choose 2× or 4×?

Use 2× when the image is already medium-sized, for example a 1200-pixel photo you want to print larger; it accepts images up to 2048 pixels. Use 4× for small images such as thumbnails, icons or old web pictures, up to 1024 pixels. Both give a result of at most 4096 pixels on the longest side.

How long does it take?

On a graphics card with WebGPU it takes a few seconds. On the processor it depends on the image size and your device: on a recent laptop a 1000 × 750 photo takes about 20 seconds when the browser can use several processor cores, and a minute or more when it can use only one; a phone can be several times slower. The progress bar shows how many parts of the image are done and an estimate of the time left.

Can I use the upscaled images commercially?

Yes, as far as this tool is concerned: the Real-ESRGAN model is released under the BSD licence, which allows commercial use, and we claim no rights in your images. You still need the rights to the original picture itself.

Upscaling uses Real-ESRGAN realesr-general-x4v3 by Xintao Wang et al. (BSD-3-Clause), in ONNX exports by techycoder and vlad-grig (BSD-3-Clause), running on ONNX Runtime Web (MIT).