AI Image Upscaler: Batch 2x, 3x, 4x Real-ESRGAN

AI Image Upscaler enlarges and sharpens images in bulk with Real-ESRGAN, a proven open-source super-resolution model. Give it a list of image links (or any Apify dataset, such as a product scraper's output) and get back 2x, 3x or 4x larger images, or images resized to an exact target size like 2048 or 3840 pixels, with transparency kept.

Unlike single-image upscalers, it processes hundreds of images per run, outputs up to 64 megapixels, and saves every result as a download link plus an optional ZIP.

✅ Pay per result: $0.017 per image ($17 per 1,000), plus Apify platform usage · 🆓 Failed and skipped images are never charged · 🔑 No API key, no GPU, images are never sent to a third-party AI service

▶ Try it with one click · 📄 Real example output · 💲 Pricing · 🔌 API & integrations · ❓ FAQ · 🛟 Support

What does AI Image Upscaler do?

• 🔍 AI super-resolution: Real-ESRGAN (realesr-general-x4v3) restores edges, texture and text instead of just blurring pixels like bicubic resizing
• 📦 Batch processing: paste many links, upload a text file of links, or read image links straight from a dataset field (dotted paths such as images.0.url work)
• 📐 2x, 3x, 4x or an exact size: set Target longest side (for example 2048 for marketplaces or 3840 for 4K) and every image is scaled to fit; images already that big are skipped for free
• 🖼️ Big outputs: up to 64 MP per image by default (for example 7200 x 4800), not a 1 to 2 MP preview
• 🫥 Transparency kept: PNG and WebP outputs keep the alpha channel; colour profiles and EXIF rotation are respected
• 💾 PNG, JPEG or WebP, adjustable quality, and an optional ZIP of all results
• 🛡️ Predictable runs: very large inputs are resized to a sensible model size first (adjustable), and the run stops cleanly at your Max cost per run

Use cases

• E-commerce teams: upscale supplier and marketplace product photos to meet minimum image sizes (1000 px+ minimums on Shopify, Etsy and other marketplaces) in bulk
• Print and merch sellers: enlarge artwork and photos for posters, mugs and print-on-demand
• Real estate and travel listings: sharpen small or compressed property and venue photos
• Content and social teams: rescue low-resolution images for blogs, thumbnails and ads
• Archives and design: enlarge scans, old photos and illustrations
• Data pipelines: chain after any scraper to upgrade every image it collected

Why choose this upscaler?

	This Actor	Typical single-image AI upscalers
Images per run	Hundreds (links, text file or dataset)	Often one per run
Output size	Up to 64 MP, 2x, 3x, 4x or exact target size	Often capped at 1 to 2 MP
Transparency (PNG/WebP alpha)	✅ kept	Often flattened
Skips images that are already big enough	✅ free	Rare
ZIP of all results	✅	Rare
Price	$0.017 per image, Silver $0.013, Gold $0.0105	$0.02 per image and up, plus usage
Failed images	Free	Varies

Honest note: Real-ESRGAN is a fast, faithful restoration model. It sharpens what is in the photo; it does not invent new detail the way slow diffusion models can.

How to use it

1. Click Try for free. The input is already filled in with two small public images (a 500 x 333 photo and a transparent 500 x 375 PNG). Press Start: it finishes in under a minute and costs about $0.04 plus a few cents of platform usage.
2. Replace the links with your own, or upload a text file with one link per line. To upscale a scraper's images, put its dataset ID in Read image links from a dataset and the field name in Dataset field with the image link.
3. Choose the Upscale factor (2x, 3x, 4x) or a Target longest side in pixels.
4. Pick the output format and, if you like, Also create a ZIP of all results.
5. Open the results table: every row has the upscaled image link, sizes and timing. Export as JSON, CSV or Excel.

Input example

{
  "images": [
    { "url": "https://thumb.wikimedia.org/wikipedia/commons/thumb/a/a7/Camponotus_flavomarginatus_ant.jpg/500px-Camponotus_flavomarginatus_ant.jpg" },
    { "url": "https://thumb.wikimedia.org/wikipedia/commons/thumb/4/47/PNG_transparency_demonstration_1.png/500px-PNG_transparency_demonstration_1.png" }
  ]
}

Other options: scale ("2", "3", "4"), targetLongestSide, skipIfAlreadyLarge, outputFormat (png, jpeg, webp), quality, createZip, datasetId, datasetImageField, maxImages, maxInputMegapixels, maxOutputMegapixels, maxDownloadMb.

Output example (from a real public run)

First row of the public example run:

{
  "sourceUrl": "https://thumb.wikimedia.org/wikipedia/commons/thumb/a/a7/Camponotus_flavomarginatus_ant.jpg/500px-Camponotus_flavomarginatus_ant.jpg",
  "inputWidth": 500,
  "inputHeight": 333,
  "inputFormat": "jpeg",
  "status": "ok",
  "charged": true,
  "outputUrl": "https://api.apify.com/v2/key-value-stores/DjDo7b6s8PJrJjlpg/records/0001-camponotus-flavomarginatus-ant-x4.png?signature=1O4XFi9ItAvXqwMrmSAtR",
  "outputKey": "0001-camponotus-flavomarginatus-ant-x4.png",
  "outputWidth": 2000,
  "outputHeight": 1332,
  "scale": 4,
  "outputFormat": "png",
  "outputMegapixels": 2.66,
  "outputSizeKb": 2075.9,
  "inputDownscaledFirst": false,
  "modelInputWidth": 500,
  "modelInputHeight": 333,
  "alphaPreserved": false,
  "model": "Real-ESRGAN realesr-general-x4v3",
  "processingSeconds": 4.6
}

Full results: JSON · CSV · Upscaled image (2000 x 1332 PNG from a 500 x 333 original).

Field	Meaning
status	ok, skipped (already big enough, free) or failed (with error, free)
outputUrl	Download link of the upscaled image
inputWidth, inputHeight, outputWidth, outputHeight, scale	Sizes in pixels and the final scale factor
modelInputWidth, modelInputHeight, inputDownscaledFirst	Size fed to the AI model (large inputs are resized to Max model input first)
alphaPreserved	Whether transparency was kept
outputSizeKb, outputMegapixels, processingSeconds	File size, image size and time taken

How much does it cost to upscale images?

Pay-per-event pricing: you pay per image successfully upscaled, the same price at every output size. Apify platform usage (compute) is billed to your account as well, like most AI image Actors.

Event	Price	Notes
Image upscaled	$0.017	Bronze $0.0155, Silver $0.013, Gold and above $0.0105
Run start	$0.0005 per GB of memory	Default 4 GB = $0.002 per run

Platform usage (measured on Apify at the default 4 GB): about 45 seconds of run time per megapixel of model input, which is roughly 0.05 compute units per megapixel. A 500 x 333 image took about 5 seconds; a 1800 x 1200 photo (resized to the default 2 MP model input, output 7200 x 4800) took 84 seconds. Apify's Free plan includes $5 of monthly credit.

Worked examples (Free/Starter event prices, default memory, usage not included):

Job	Calculation	Event charges
One-click example (2 small images)	2 × $0.017 + $0.002 start	$0.036
1,000 product photos to 2048 px	1,000 × $0.017 + $0.002	$17.00
Same job on the Gold plan	1,000 × $0.0105 + $0.002	$10.50
300 images, 120 already over the target (skipped)	180 × $0.017 + $0.002	$3.06

Never charged: images that fail (404, not an image, web page instead of an image, too big to download) and images skipped because they already reach your target size. Cap spending with Max cost per run: the Actor stops cleanly at the limit.

Integrations (API, AI agents, Make, Zapier)

API: upscale and get results in one call:

curl -X POST "https://api.apify.com/v2/acts/chris_elder~ai-image-upscaler/run-sync-get-dataset-items" \
  -H "Authorization: Bearer $APIFY_TOKEN" -H "Content-Type: application/json" \
  -d '{"images":[{"url":"https://thumb.wikimedia.org/wikipedia/commons/thumb/a/a7/Camponotus_flavomarginatus_ant.jpg/500px-Camponotus_flavomarginatus_ant.jpg"}],"scale":"4"}'

from apify_client import ApifyClient
client = ApifyClient("<YOUR_APIFY_TOKEN>")
# Upscale every image collected by a previous scraper run
run = client.actor("chris_elder/ai-image-upscaler").call(run_input={
    "datasetId": "<SCRAPER_DATASET_ID>", "datasetImageField": "imageUrl",
    "targetLongestSide": 2048, "outputFormat": "jpeg", "createZip": True,
})
for row in client.dataset(run["defaultDatasetId"]).iterate_items():
    print(row["status"], row.get("outputUrl"))

AI agents (MCP): add https://mcp.apify.com/?tools=chris_elder/ai-image-upscaler to Claude, Cursor, VS Code or any MCP client, and your agent can upscale image links on request.

Make, Zapier, n8n: use the Apify app's Run Actor action with chris_elder/ai-image-upscaler, then send the image links to Google Drive, Dropbox, Shopify or your CMS. Webhooks fire when a run finishes, so you can chain it after any scraper.

Limits and good to know

• Real-ESRGAN works best on photos, product shots, illustrations and screenshots with mild blur or JPEG artefacts. Heavily damaged or tiny faces will not be reconstructed like a dedicated face restorer would.
• The model upscales 4x natively. 2x and 3x are made from the 4x result; scales above 4x (for example a 500 px image to 3840 px) use the 4x AI pass plus high-quality Lanczos resizing for the rest.
• Max model input defaults to 2 MP (up to 8): larger originals are resized to that size before the AI pass, then upscaled. Raise it for more detail on big photos; runs take longer.
• Output is capped at 64 MP by default (adjustable to 100 MP). JPEG output turns transparent areas white.
• Images are processed one at a time on CPU (the next one downloads in the background). More memory gives more CPU and faster runs.
• Animated GIFs are upscaled as their first frame. Downloads over 50 MB are skipped by default.

Is it legal?

The Actor downloads only the image links you give it (or the links in your dataset), one request per image, and processes them on Apify's servers. It doesn't crawl websites, log in or bypass protection. Make sure you have the right to modify and reuse the images (your own photos, supplier images you are licensed to use, public-domain or openly licensed images). See Apify's guide: Is web scraping legal?

FAQ

How is this different from resizing in an image editor? Normal resizing interpolates pixels, so enlarged images look soft. Real-ESRGAN is trained to restore sharp edges and texture and to clean JPEG artefacts while enlarging.

Can I upscale the images from another scraper? Yes. Put that run's dataset ID in Read image links from a dataset and the field that holds the image URL (for example imageUrl, images or media.0.url). Lists of links in one field are all processed.

How big can the output be? Up to 64 MP by default, for example 8000 x 8000 or 9200 x 6900. Set Target longest side for an exact size.

Do I pay for images that are already large enough? No. With a target size set and Skip images that already reach the target on (the default), those images get a skipped row and are not charged.

Is my image sent to an external AI service? No. The open-source model runs inside the Actor on Apify.

What happens if an image fails? You get a row with status: "failed" and the reason, and you are not charged for it. The run carries on.

Support, reliability and feature requests

• Failed images are free. You pay only for images upscaled.
• Tested on real images before every release (small and large JPEGs, transparent PNGs, a 28 MP scan, exact-size targets, broken links and web pages).
• Issues answered within 2 business days. Open an issue on the Issues tab with a sample image link.
• Feature requests welcome: open an issue titled "Feature request: ...".
• Built by GPT Shop Ltd with AI-assisted development; every release is reviewed and tested before it ships.

More tools from GPT Shop Ltd

• Audio & Video Transcriber: Whisper speech to text with SRT and VTT subtitles, podcast feeds included
• Bulk File Downloader: download many files or images to storage or a ZIP
• Image to Text OCR: bulk text extraction from images in 100+ languages
• Bulk Website Screenshot: full-page screenshots of many URLs
• All tools: apify.com/chris_elder