AI Image Upscaler - Real-ESRGAN 4x for Photos, Products & Anime

Enlarge images 2x or 4x with AI and get back sharper detail, cleaner edges and less blur and JPEG noise than any normal resize. Paste image links, get public download links to the upscaled files. Transparency is kept, and there are two models: one for photos and real-world images, one for anime, cartoons, illustrations and line art.

It runs Real-ESRGAN, the widely used open-source super-resolution model (BSD-3 licence), inside the Actor - no third-party AI API, no account, no watermark, no subscription.

From $0.01 per image, and failed images are free.

What does AI Image Upscaler do?

AI Image Upscaler downloads each image you link to, enlarges it 2x or 4x with the Real-ESRGAN neural network (a model for photos and one for anime, cartoons and line art), keeps transparency, and saves the result as JPEG, PNG or WEBP with a public download link. You get one dataset row per image with the input and output sizes.

Use cases

• Ecommerce - turn small supplier product photos into sharp 2000+ px images for your store and marketplaces.
• Print and merch - enlarge artwork and photos to print resolution.
• Old and low-res photos - make thumbnails, scans and screenshots presentable again.
• Anime, game and illustration art - the anime model keeps flat colours and clean lines.
• Before OCR or computer vision - upscale tiny text and small images first (pairs with Image to Text OCR).
• Bulk pipelines - collect images with Bulk Image Downloader, then upscale them all here; or call it from Make, Zapier, n8n, the API or an AI agent via MCP.

Who is it for?

• Ecommerce sellers turning small supplier photos into sharp store images.
• Print-on-demand and merch creators enlarging artwork to print size.
• Designers and marketers rescuing low-resolution assets and old photos.
• Anime, game and illustration artists who need clean lines kept.
• Developers who need an image upscaling API without running a GPU.

How to use it

1. Click Try for free.
2. Paste direct image links into Image URLs, one per line.
3. Pick Scale (4x or 2x) and Model (general for photos, anime for drawn art).
4. Click Start, then open or download the upscaled images from the result links.

Input

Field	What it does	Default
imageUrls	Direct image links, one per line (JPEG, PNG, WEBP, GIF first frame, BMP, TIFF; up to 40 MB)	two sample images
scale	4 or 2	4
model	general (photos) or anime (illustrations, cartoons, line art)	general
outputFormat	jpeg, png (lossless) or webp	jpeg
quality	JPEG / WEBP quality, 50-100	95
allowLargeImages	Upscale inputs up to ~2.1 MP (1920x1080) at full size, at the large price	false

{
    "imageUrls": ["https://example.com/product-small.jpg", "https://example.com/logo.png"],
    "scale": "4",
    "model": "general",
    "outputFormat": "png"
}

Image sizes. The standard price covers inputs up to about 0.3 megapixels (640x480, 800x400, 550x550...), which come out at up to 2560x1920 at 4x. Bigger inputs are first reduced to that size

• unless you turn on allowLargeImages, which upscales inputs up to ~2.1 MP (1920x1080) at full size (up to 7680x4320 at 4x) at the large price. An image that is already bigger than the result would be is refused with an explanation, free of charge.

Output

Each image is saved in the run's key-value store and gets a public link. Real output from a cloud run:

{
    "imageUrl": "https://upload.wikimedia.org/wikipedia/commons/thumb/3/3f/Fronalpstock_big.jpg/960px-Fronalpstock_big.jpg",
    "outputUrl": "https://api.apify.com/v2/key-value-stores/2d3YTKxifpnlGRTca/records/0000-960px-Fronalpstock_big-x4.jpg?signature=FqikHLWogeqP3wVIxVUZ",
    "outputKey": "0000-960px-Fronalpstock_big-x4.jpg",
    "model": "general",
    "scale": 4,
    "inputFormat": "JPEG",
    "width": 960,
    "height": 431,
    "networkInputWidth": 960,
    "networkInputHeight": 431,
    "inputReduced": false,
    "outputWidth": 3840,
    "outputHeight": 1724,
    "effectiveScale": 4,
    "tier": "large",
    "hasTransparency": false,
    "inferenceMs": 15684,
    "outputFormat": "jpeg",
    "chargedAs": "large-image-upscaled",
    "error": null
}

A transparent PNG with the anime model at 2x came back as a 500x374 RGBA PNG with its alpha channel intact. A failed download is reported like this, and not charged:

{ "imageUrl": "https://example.com/missing.jpg", "outputUrl": null, "error": "HTTPStatusError: Client error '404 Not Found' for url 'https://example.com/missing.jpg'" }

The Console table shows the original and the upscaled image side by side.

How much does it cost to upscale an image?

Pay per event, only for images that were upscaled and saved:

Event	When	Price
Image upscaled	input up to ~0.3 MP (output up to 2560x1920 at 4x)	$0.01
Large image upscaled	input 0.3-2.1 MP, only with allowLargeImages	$0.06

Plus a $0.00005 start fee. 100 product photos at the standard size cost $1.00. Set "Maximum cost per run" and the run stops cleanly when it is reached.

Speed on the default 4 GB memory: about 12 seconds for a standard image and about 80 seconds for a full 1920x1080 input. Images download in parallel while one is being upscaled.

Integrations and API

Run it from your own code with the official apify-client package (Python shown; JavaScript works the same way):

from apify_client import ApifyClient

client = ApifyClient("YOUR_APIFY_TOKEN")
run = client.actor("kantolabs/ai-image-upscaler").call(run_input={
    "imageUrls": [
        "https://upload.wikimedia.org/wikipedia/commons/thumb/3/3f/Fronalpstock_big.jpg/330px-Fronalpstock_big.jpg"
    ],
    "scale": "4",
    "model": "general"
})
for item in client.dataset(run["defaultDatasetId"]).iterate_items():
    print(item)

Or get the results in a single HTTP call:

curl -X POST "https://api.apify.com/v2/acts/kantolabs~ai-image-upscaler/run-sync-get-dataset-items?token=YOUR_APIFY_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"imageUrls": ["https://upload.wikimedia.org/wikipedia/commons/thumb/3/3f/Fronalpstock_big.jpg/330px-Fronalpstock_big.jpg"], "scale": "4", "model": "general"}'

• AI agents (MCP): add the Apify MCP server to Claude, ChatGPT, Cursor or any MCP client and the agent can call this Actor as a tool.
• No-code: Make, Zapier, n8n, Google Sheets, webhooks and Apify Schedules work out of the box.

Limits

• Runs on CPU: about 12 seconds per standard image and about 80 seconds per full 1920x1080 input.
• The standard price covers inputs up to ~0.3 MP; larger inputs are reduced first unless allowLargeImages is on (up to ~2.1 MP at full size).
• Up to 40 MB per file. Links from social networks and stock-photo sites are refused on purpose.
• It sharpens what is there; it does not restore faces or invent text that was never in the image.

FAQ

How good is it? Real-ESRGAN is the reference open-source upscaler used by many desktop and web tools. It recovers edges and textures and removes compression noise; it does not invent faces or text that were never there. Tiny text and faces benefit less than landscapes, products and art.

Which model should I pick? general for photos, product shots, screenshots and scans. anime for anything drawn: anime, cartoons, comics, game art, logos and line art.

Is my image stored anywhere else? No. It is downloaded by the run, processed inside the Actor and saved only in your run's storage.

Which sites can I use? Any direct image link you have the right to process. Links from social networks and stock-photo sites are refused on purpose.

Related Actors by Kanto Labs

• Bulk Image Downloader - pages or URLs to ZIP & links
• Image to Text OCR - photos, screenshots, scanned PDFs
• Video Frame Extractor - keyframes, scenes & thumbnails