AI Image Upscaler 2x 3x 4x: Real-ESRGAN Super Resolution No GPU

Make small photos and images bigger and sharper. Give it image links, choose 2x, 3x or 4x, and get the upscaled files back as PNG, JPEG or WebP, with one row per image that links to the file.

Paste a few image links and press Start.

What people use it for

• Product and listing photos: bring small supplier images up to the size a store or marketplace asks for.
• Old or low resolution pictures: make them large enough to print or to use on a wide screen.
• Pipelines and AI agents: call it from the API or an integration whenever an image is too small.

How it works

It runs the open Real-ESRGAN general model (realesr-general-x4v3) on CPU. The model sharpens edges and removes some blur and JPEG noise while it enlarges. Then the result is resized to the exact scale you chose.

Transparency in PNG and WebP images is kept. Photos are turned the right way up from their EXIF orientation first.

What it does not do

• It does not invent detail that is not there. Faces, small text and fine patterns get smoother and sharper, not restored. Very small or heavily compressed images can look painted.
• It is not a face restoration or generative tool, and it does not change the content of the image.
• Animated GIFs come back as one still frame.
• Each image takes a few seconds per output megapixel, so very large outputs are slower.

Input

Field	What it does
Image URLs	Direct links to JPEG, PNG, WebP, BMP, TIFF or GIF images, or files you upload. Sites whose robots.txt disallows automated access are skipped.
Scale	2x, 3x or 4x the width and height.
Output format	PNG keeps every pixel and transparency. JPEG and WebP are much smaller files.
JPEG and WebP quality	From 50 to 100.
Max output megapixels	Outputs larger than this are scaled down to fit, so one huge image cannot run up the bill. Default 16.
Max input file size (MB)	Larger files are skipped.

Output

One item per upscaled image. The file itself is saved in the run's key-value store, and imageUrl links to it.

{
  "sourceUrl": "https://plainrecordlabs.web.app/sample-apollo11-320.jpg",
  "imageKey": "0001-sample-apollo11-320-upscaled.png",
  "imageUrl": "https://api.apify.com/v2/key-value-stores/<store id>/records/0001-sample-apollo11-320-upscaled.png?signature=<signature>",
  "inputWidth": 320,
  "inputHeight": 312,
  "outputWidth": 1280,
  "outputHeight": 1248,
  "outputMegapixels": 1.6,
  "megapixelsCharged": 2,
  "scale": 4,
  "format": "png",
  "fileBytes": 2412345,
  "seconds": 3.1
}

The example values are illustrative. Images that could not be upscaled are not charged and are listed with the reason in the run's SUMMARY record.

Privacy

Your images are fetched only for your run and processed in memory. The results stay in your own run's storage and nothing is kept anywhere else.

FAQ

How do I upscale an image online without a GPU? Paste the image link or upload a file, choose 2x, 3x or 4x and press Start. The model runs on CPU on Apify, so you need no graphics card or software of your own.

Which AI upscaling model does it use? The open Real-ESRGAN general model (realesr-general-x4v3).

Can I upscale many images at once? Yes. Add as many image links as you like. Each image comes back as its own row with a link to the file.

Will it restore faces or blurry photos? No. It sharpens and enlarges but does not invent detail, so faces and small text get smoother, not restored.

Which formats does it support? It reads JPEG, PNG, WebP, BMP, TIFF and GIF, and saves PNG, JPEG or WebP. PNG and WebP keep transparency.

Pricing

You pay a small fee per image plus a fee per started megapixel of output. A 320 x 312 photo at 4x is 1.6 megapixels, so it counts as 2. The prices are on the Pricing tab. Max output megapixels caps what one image can cost.

Credits

The model is Real-ESRGAN by Xintao Wang and others, released under the BSD 3-Clause License. The licence text ships with this Actor in model/LICENSE-Real-ESRGAN.txt. The sample photo is a public domain NASA image from Apollo 11.

Feedback

Need another option or a fix? Open an issue on the Issues tab.