AI Image Upscaler – Real-ESRGAN 2x/4x Photo Enhancer

Upscale photos, product images, artwork and old pictures 2x, 3x or 4x with Real-ESRGAN, an open-source AI super-resolution model that restores detail instead of just blurring pixels. Paste image URLs, click Start, and download sharp, high-resolution PNG, JPEG or WebP files. You can process bulk batches through the API, with no subscription and no third-party AI API fees.

Why use it

• Real AI super-resolution. It uses Real-ESRGAN (realesr-general-x4v3) and removes JPEG artifacts, noise and blur while it enlarges.
• Two models. General gives the sharpest results. Denoise is gentler and suits noisy phone photos and old scans.
• 2x, 3x or 4x output, with transparency (PNG alpha) preserved.
• Bulk and API-first. Process hundreds of images per run and call it from Make, Zapier, n8n or your own code.
• Private. Images are processed on Apify's servers inside your run. Nothing is sent to an external AI service.
• Fair pricing. You pay per image, and large images are billed per started megapixel of input. Failed downloads are free.

How to use it

1. Add image URLs under Images. You can also upload a text file of URLs. For local files, upload them to an Apify key-value store, Dropbox, Google Drive (public link) or S3.
2. Choose the Scale (4x by default), the Model and the Output format.
3. Click Start. Each result has an outputUrl that links to the upscaled file in your run's key-value store.

Input example

{
  "images": [{ "url": "https://raw.githubusercontent.com/xinntao/Real-ESRGAN/master/inputs/wolf_gray.jpg" }],
  "scale": "4",
  "model": "general",
  "outputFormat": "png"
}

Output example

{
  "url": "https://raw.githubusercontent.com/xinntao/Real-ESRGAN/master/inputs/wolf_gray.jpg",
  "status": "OK",
  "outputUrl": "https://api.apify.com/v2/key-value-stores/…/records/wolf_gray-1-x4.png",
  "scale": 4,
  "model": "general",
  "originalWidth": 236,
  "originalHeight": 236,
  "outputWidth": 944,
  "outputHeight": 944,
  "outputFormat": "png",
  "outputBytes": 1003760,
  "billedImages": 1,
  "processingSeconds": 1.4
}

status values: OK, FAILED, HTTP_ERROR, SKIPPED_ROBOTS_TXT. Only OK results are charged.

Pricing (pay per event)

Event	Price
Image upscaled (per started megapixel of input, i.e. up to 1024×1024 px)	$0.02

Examples: a 512×512 thumbnail costs $0.02 and comes back as 2048×2048 at 4x. A 1600×1200 photo (1.83 MP) counts as 2 units and costs $0.04. 1,000 product images under 1 MP cost $20. Inputs larger than Max input megapixels (2 MP by default) are first downscaled to that size, which keeps runs fast and cheap. Raise the limit for big originals.

FAQ

What's the maximum output size? With the default 2 MP input limit at 4x, outputs reach about 32 MP (for example 6400×5000). You can raise the input limit to 8 MP.

Does it work on faces and text? It works well on general photos, products, landscapes and illustrations. It doesn't hallucinate new faces, so very low-res faces get sharper but aren't "reconstructed".

Is 2x cheaper than 4x? No. The price depends on input size, because the AI model runs at 4x and the result is then resized to your chosen scale for the best quality.

Which formats are supported? Input can be JPEG, PNG, WebP, BMP, TIFF or GIF (first frame). Output can be PNG, JPEG or WebP.

Can I use the results commercially? Real-ESRGAN is BSD-3 licensed, and you own your outputs. Make sure you have the rights to the input images.