AI Image Upscaler & Photo Restoration – Face Enhance, 2x / 3x / 4x

Upscale, sharpen and restore photos in bulk: Real-ESRGAN super-resolution for the whole image plus GFPGAN AI face restoration for portraits, old family photos and blurry selfies. Paste image URLs (or chain a dataset from any scraper) and get 2x, 3x or 4x larger, cleaner images back as JPG, PNG or WebP files with direct download links.

• Old, blurry or low-res photos: the general model (realesr-general-x4v3) restores detail and removes JPEG artefacts and scan noise, with an adjustable denoise strength.
• Face restoration (new): faces are detected automatically and rebuilt with GFPGAN v1.4: sharper eyes, skin, teeth and hair. Black-and-white photos stay black-and-white. Images without faces cost nothing extra.
• Anime, cartoons, illustrations, logos: the anime model keeps lines crisp and colours flat.
• Transparency kept: PNG/WebP with alpha stay transparent.
• Bulk & automation friendly: lists, CSV, Google Sheets, or datasetId from another Actor; API, schedules, webhooks, Make/Zapier/n8n.
• Pay per image, no GPU, no subscription, no desktop app: an API alternative to desktop photo-enhancer tools. Failed downloads and invalid files are never charged.

What it costs

Event	Price	When
image-upscaled	$0.005	per successfully upscaled image, includes up to 0.25 MP of input (e.g. 500×500 px → 2000×2000 output)
extra-input-block	$0.003	per additional started 0.25 MP of input
face-restored	$0.008	per face restored (only when faceEnhance is on and a face was found; up to maxFaces per image)

Examples: 400×400 thumbnail → $0.005 · old 700×460 portrait with 2 faces → $0.005 + $0.003 (1 extra block) + $0.016 (2 faces) = $0.024 · 800×600 photo (0.48 MP) → $0.008 · 1000×1000 (1 MP) → $0.014 · 1920×1080 (2.07 MP) → $0.029. Errors and skipped images are free. Set Max input size to guard your budget; the run also stops cleanly at your maximum cost per run.

Input

Field	Default	Notes
imageUrls	–	Direct image links (JPEG, PNG, WebP, HEIC, AVIF, TIFF, GIF, BMP)
bulkText / sourceFileUrl / datasetId	–	Bulk sources (one URL per line, CSV with a url column, or another dataset)
scale	4	2, 3 or 4. 2x/3x are downsampled from the 4x result
model	general	general (photos) or anime (illustrations)
denoiseStrength	0.5	0 = keep grain/texture, 1 = strongest noise & artefact removal (general only)
faceEnhance	true	Detect and restore faces with GFPGAN (charged per restored face)
faceEnhanceStrength	0.8	1 = full restoration, lower = keep more of the original look
maxFaces	5	Faces restored per image, largest first
targetWidth / targetHeight	–	Exact output size instead of a factor (aspect kept, max 4x)
outputFormat	jpg	jpg, png or webp; quality 50–100
maxInputMegapixels	4	Bigger inputs are skipped for free
maxOutputMegapixels	32	Factor is reduced to stay below this

Example:

{
  "imageUrls": ["https://example.com/product-small.jpg"],
  "scale": "4",
  "model": "general",
  "denoiseStrength": 0.5,
  "outputFormat": "jpg"
}

Output

Every image becomes one dataset row; the file itself is stored in the run's key-value store and linked in outputUrl.

{
  "url": "https://example.com/product-small.jpg",
  "status": "ok",
  "outputUrl": "https://api.apify.com/v2/key-value-stores/<id>/records/upscaled-0001-product-small-1a2b3c.jpg",
  "inputWidth": 500, "inputHeight": 500,
  "outputWidth": 2000, "outputHeight": 2000, "outputMegapixels": 4.0,
  "scale": 4, "model": "general", "denoiseStrength": 0.5,
  "format": "jpg", "outputBytes": 812345,
  "facesDetected": 1, "facesRestored": 1,
  "chargedBlocks": 1, "processingSeconds": 6.1
}

Errors look like {"url": "...", "status": "error", "errorCode": "HTTP_404", "error": "HTTP 404 (not charged)"}. Codes: INVALID_URL, HTTP_4xx/5xx, TIMEOUT, TOO_LARGE, INPUT_TOO_LARGE, NOT_AN_IMAGE, DECODE_ERROR, OUT_OF_MEMORY.

Use cases

• E-commerce: enlarge supplier/product photos for Shopify, Amazon, Etsy zoom requirements (e.g. 2000 px minimum).
• Print on demand & Etsy sellers: upscale artwork to print resolution.
• Real estate & marketplaces: sharpen small listing photos.
• Old family photos & genealogy: restore scanned prints and faded portraits before printing or sharing.
• Profile pictures & headshots: sharpen small or blurry face photos.
• Content & social media: rescue low-res images, screenshots and memes.
• Anime & game art: clean 4x upscales of illustrations and sprites.
• Pipelines: scrape images with any Actor, then pass the dataset to this one.

Speed & memory

Runs on CPU. With the default 2 GB a 500×500 image takes roughly 10–15 s, plus about 5–9 s per restored face; 4 GB is about twice as fast, 8 GB faster still. Memory does not change the price per image. Very large outputs (>32 MP) need 8 GB+.

Limits & honest notes

• Upscaling invents plausible detail; it cannot recover text that is not in the original. Face restoration generates a plausible sharp face from the blurry one: very small or heavily damaged faces can look slightly different from the real person. Lower faceEnhanceStrength (or turn faceEnhance off) when exact likeness matters.
• Only process images you have the right to use. Downloads use a descriptive User-Agent.
• Platform pages (Instagram, Google Images, etc.) are not image files: pass direct image URLs.

Credits

Models: Real-ESRGAN by Xintao Wang et al. (BSD-3-Clause); GFPGAN v1.4 by Tencent ARC (Apache-2.0, clean architecture without the NVIDIA CUDA ops); YuNet face detector (MIT). All converted to ONNX and run with ONNX Runtime. This Actor is not affiliated with these authors.

Related Actors by gazidev

Image Downloader · Image Metadata & EXIF Extractor · Image OCR · Video Frame Extractor · Website Screenshot