AI Image Upscaler x4 – Anime & Photo (Real-ESRGAN)

Make small images 4x larger (or 2x) and sharper with Real-ESRGAN, the widely used open-source super-resolution model. Pick the model that fits your images:

• 🎨 Anime / illustration – Real-ESRGAN's anime model: clean line art and flat colors for anime, manga, illustrations, game sprites and stickers.
• 📷 Photo / general – Real-ESRGAN's general model for photos, product shots and real-world images. Fast.
• 🧩 Up to 2 MP input – images are processed in overlapping tiles, so seams are not visible and memory stays bounded (about 2.4 GB at the 2 MP limit with the anime model, within the default 4 GB).
• 🪟 Transparency kept – PNG and WebP output keep the alpha channel.
• 💸 Fair pricing – pay per 512×512 block of the input image; failed images are free.
• 🤖 API-ready – one dataset item per image with a direct link to the result.

How to use

1. Paste direct image links into Image URLs (PNG, JPEG, WebP, GIF, BMP, TIFF; up to 100 per run).
2. Choose Model: anime or photo.
3. Optional: Scale 4x (default) or 2x, Output format PNG / JPG / WebP.
4. Run. Each result is saved in the run's key-value store and linked from the dataset item.

Leave Image URLs empty to try a free bundled sample.

Pricing

Billed per tile = each started 512×512 block of the input image (262,144 input pixels):

Model	Price per tile	Examples
Photo / general	$0.01	512×512 → 2048×2048: $0.01 · 1024×768 → 4096×3072: $0.03
Anime / illustration	$0.05	512×512 → 2048×2048: $0.05 · 800×600 → 3200×2400: $0.10

Failed images (bad link, not an image, too large) are not charged. Each run also has Apify's standard start fee: $0.00005 per run up to 1 GB of memory, plus $0.00005 for each extra GB. Set Max total charge to cap your spend – images that would exceed it are skipped and reported.

More memory = more CPU = faster runs: 4 GB (default) uses 1 CPU core, 8 GB uses 2, 16 GB uses 4. The per-tile price is the same; only the small start fee grows with memory.

Input

Field	Default	Notes
imageUrls	–	Up to 100 per run; 30 MB and 2 megapixels (~1414×1414) each
model	photo	photo or anime
scale	4	4 or 2 (x2 = model at x4, then high-quality resize)
outputFormat	png	png, jpg, webp
quality	92	JPG / WebP quality, 50–100

Output

{
  "url": "https://example.com/small.png",
  "status": "ok",
  "model": "anime",
  "scale": 4,
  "inputWidth": 512, "inputHeight": 512,
  "outputWidth": 2048, "outputHeight": 2048,
  "tiles": 1,
  "bytes": 3104422,
  "imageUrl": "https://api.apify.com/v2/key-value-stores/…/records/image-0.png"
}

status is ok, error (with message, not charged) or skipped (max total charge reached, not charged).

FAQ

Which model should I use? Drawn images with outlines and flat colors → anime. Photos and anything with natural texture → photo.

Will it invent details? Like every AI upscaler it reconstructs plausible edges and textures; it does not add new objects. Small text and faces in very low-resolution images may not be recovered exactly.

Are larger inputs supported? Up to 2 megapixels per image. For bigger images, crop or resize first.

How is my data handled? Images are downloaded into your run, upscaled on Apify's CPU (no external AI service), and saved only in your Apify storage under Apify's retention settings. We keep no copies. Only process images you have the right to use.

Notes

• Models: Real-ESRGAN realesr-general-x4v3 and RealESRGAN_x4plus_anime_6B by Xintao Wang et al., BSD 3-Clause License, official release weights, unmodified. Network code follows Real-ESRGAN (BSD-3) and BasicSR (Apache 2.0). Full notices: THIRD_PARTY_NOTICES.md.
• The bundled sample image is our own drawing.
• Not affiliated with the Real-ESRGAN authors or Tencent ARC.