Fast Image Upscaler 2x/4x for Graphics, Logos, AI Art & Product Shots

Upscale batches of graphics, icons, logos, clip art, AI art, 3D renders and simple product shots by 2x or 4x with the compact Real-ESRGAN model. Paste image URLs and get sharp, clean PNG, JPEG or WebP files back, plus one result row per image.

• Cheap and predictable: $0.009 per upscaled image, same price for 2x and 4x. Failed, oversized, blocked or timed-out images are free.
• Clean edges: removes JPEG blockiness and blur on flat colours, outlines, icons and illustrations.
• Keeps transparency: transparent PNG/WebP logos and icons stay transparent.
• Batch friendly: up to 100 images per run, with a clear per-image status and error message.
• Safe limits: per-image size, pixel and time caps, so a run can't blow up your bill.

⚠️ Not for faces or portraits

This is a small, fast model. On faces, skin, hair and fine natural textures it gives a smooth, waxy, "painted" look. For photos of people, use a heavier photo or face-restoration upscaler instead. It also can't recover very small text: big lettering gets crisper, but tiny print may come out garbled (see the packaging example below).

Before / after examples

Real outputs of this Actor on the Apify platform. Left: the original enlarged 4x with a normal bicubic resize. Right: this Actor's 4x output. Same 440×440 crop at full output size. All source images are CC0 (public domain) from Wikimedia Commons. Credits and sources: CREDITS.md.

Type	Before vs after (100% crop)
Flat icon, transparent PNG, 512×512 → 2048×2048
3D render clip art, 500×281 → 2000×1124
AI art (Stable Diffusion), 512×512 → 2048×2048
AI art, 330×330 → 1320×1320
Product shot, 330×259 → 1320×1036
Product packaging, 500×371 → 2000×1484. Big text sharper, small text garbled
Weak spot: natural textures (a rough wall turns into an artificial pattern)

Input

{
  "imageUrls": [
    "https://upload.wikimedia.org/wikipedia/commons/1/1b/Online-icon.png",
    "https://example.com/my-logo.png"
  ],
  "scale": "4",
  "outputFormat": "png"
}

Field	Default	Notes
imageUrls	one small CC0 icon	Public http(s) links to PNG, JPEG, WebP, GIF (first frame), BMP or TIFF.
scale	"4"	"4" or "2". Same price.
outputFormat	png	png (lossless, keeps transparency), jpeg (transparent areas on white), webp (keeps transparency, small files).
jpegQuality	92	For JPEG and WebP.
maxImages	20	Up to 100 per run. Extra URLs are listed as skipped (free).
maxFileSizeMb	10	Up to 15 MB per image.
perImageTimeoutSecs	120	Up to 300 s per image.

The Actor also respects your maximum cost per run: once it's reached, the remaining images are listed as skipped and not processed.

Limits (hard caps)

Limit	Value
Images per run	100
File size per image	15 MB
Input size	0.26 megapixels (e.g. 512×512, 640×400, 724×362), longest side ≤ 2048 px, shortest ≥ 8 px
Output size	4.2 megapixels (e.g. 2048×2048)
Time per image	300 s

Bigger images get a clear input-too-large error (free). Resize them first. These caps keep each image fast and the price fixed.

Output

One dataset row per image. Upscaled files are saved in the run's key-value store, and outputUrl is a direct link.

{
  "index": 1,
  "sourceUrl": "https://upload.wikimedia.org/wikipedia/commons/1/1b/Online-icon.png",
  "status": "ok",
  "scale": 4,
  "inputWidth": 256, "inputHeight": 256, "inputFormat": "PNG",
  "outputWidth": 1024, "outputHeight": 1024, "outputMegapixels": 1.049,
  "outputFormat": "png", "outputBytes": 605596,
  "outputKey": "upscaled-001-Online-icon.png",
  "outputUrl": "https://api.apify.com/v2/key-value-stores/<store>/records/upscaled-001-Online-icon.png?signature=…",
  "downloadSecs": 0.08, "upscaleSecs": 1.55, "encodeSaveSecs": 0.34, "totalSecs": 2.0
}

Failed images have status: "failed" (or "skipped"), an errorCode and a plain-English error, for example input-too-large, file-too-large, not-an-image, download-failed, download-blocked, blocked-url, invalid-url, timeout or charge-limit-reached. They are never charged.

Speed and memory

The default is 1 GB, the cheapest setting for small runs.

💡 Tip for batches: set Memory = 4 GB in the run options. In our tests on Apify, batch upscaling at 4 GB was about 5x faster and used about 25–40% less platform compute per image than at 1 GB:

• At 1 GB: the run gets a full CPU burst only for the first 20–40 s. After that it drops to about a quarter of a core, around 11–15 s per output megapixel.
• At 4 GB: it stays at 2–3 s per output megapixel.

Your price stays $0.009 per image either way: under pay-per-event, the compute is included. Peak memory in our tests stayed under 600 MB.

Pricing

Pay per event: $0.009 per successfully upscaled image ($9 per 1,000). Nothing is charged for images that fail or are skipped.

How it works

• Model: Real-ESRGAN realesr-general-x4v3 (compact SRVGG network, 4x), run on CPU with ONNX Runtime. The image is processed in overlapping tiles, so memory stays flat.
• 2x = the model's 4x output, downscaled with a high-quality Lanczos filter. This is Real-ESRGAN's own "outscale" method.
• Transparency is upscaled separately with Lanczos and re-attached.
• Only public http(s) URLs are fetched. Private and internal addresses are blocked, including through redirects. Each download is size-limited while it streams.

Licences and attribution

This Actor uses the Real-ESRGAN model and weights by Xintao Wang, Liangbin Xie, Chao Dong and Ying Shan (github.com/xinntao/Real-ESRGAN). The weights realesr-general-x4v3.pth come from the project's official v0.2.5.0 release, SHA-256 8dc7edb9…6292. They are shipped as an unmodified ONNX export. Full details are in licenses/NOTICE.md, and the licence text is in licenses/Real-ESRGAN-LICENSE.txt:

BSD 3-Clause License

Copyright (c) 2021, Xintao Wang
All rights reserved.

Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are met:

1. Redistributions of source code must retain the above copyright notice, this
   list of conditions and the following disclaimer.

2. Redistributions in binary form must reproduce the above copyright notice,
   this list of conditions and the following disclaimer in the documentation
   and/or other materials provided with the distribution.

3. Neither the name of the copyright holder nor the names of its
   contributors may be used to endorse or promote products derived from
   this software without specific prior written permission.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

Citation: Xintao Wang, Liangbin Xie, Chao Dong, Ying Shan. Real-ESRGAN: Training Real-World Blind Super-Resolution with Pure Synthetic Data. ICCV Workshops, 2021.

Make sure you have the right to upscale and use the images you submit. This Actor doesn't store your images beyond the run's own storage.