AppWatch outofficial publisher

Real-ESRGAN

App : A product you use directly

Local image and video upscaling, the backbone of many tools.

The reference open-source upscaling model.

Official site GitHub · 37,209 ★

Overall score

6.6 /10

Security: nothing to flag

Utility

8.0

Reliability

2.0

Security

No caveat on record

Ease

Technical · Technical

Security reflects where data goes, requested permissions and incidents on record at the verification date. It is not a guarantee of zero risk: check the deployment mode and the connections you enable.

What it does, concretely

Old or low-resolution photos and videos lose detail as soon as they are enlarged with conventional tools. Real-ESRGAN addresses this by reconstructing missing details rather than simply interpolating pixels.

The tool relies on a neural network trained exclusively on synthetic data, allowing it to handle generally degraded images without needing real-world paired examples. It offers several models depending on the content: real-world photos, anime illustrations, and animation videos.

Real-ESRGAN is aimed at photographers, graphic designers, retouching enthusiasts, and developers who want to upscale images or restore videos without sending their files to a third-party server. All processing happens locally on the user's machine.

However, the tool is not designed for beginners: installation requires Python, PyTorch, and several dependencies, or the use of a portable command-line executable. There is no official built-in graphical interface.

What it offers

·Image upscaling up to x4 for general photos and illustrations.
·Dedicated model for anime illustrations (RealESRGAN_x4plus_anime_6B), lighter and optimised for this type of content.
·Dedicated model for animation videos (AnimeVideo-v3) to restore animated sequences.
·Face enhancement option via GFPGAN integration, enabled with the --face_enhance argument.
·Portable executables available for Windows, Linux, and macOS, compatible with Intel, AMD, and Nvidia GPUs without a Python installation.
·Adjustable denoising strength (-dn) on the realesr-general-x4v3 model to avoid over-smoothing.
·Tile-based processing to handle large images on machines with limited GPU memory.
·Support for images with alpha channel, grayscale images, and 16-bit images.

Our tips

For anime illustrations, use the RealESRGAN_x4plus_anime_6B model rather than the general one: it is lighter and produces better results on this type of content.
If the result looks too smooth, use the realesr-general-x4v3 model with the -dn option to adjust the denoising strength.
On a machine with limited GPU memory, enable the --tile option with a value such as 256 or 512 to split the image into chunks and avoid memory errors.
The --outscale argument allows you to obtain an upscaling factor other than x4: the model processes at x4 and then resizes the result using a standard algorithm (LANCZOS4).
To automatically enhance faces in an image, add --face_enhance to the Python command, which activates GFPGAN alongside Real-ESRGAN.

Watch out

The portable executables (NCNN) split the image into tiles before stitching them back together, which can introduce visible inconsistencies at the seams and produce slightly different results from the Python version.
Installation via Python requires managing several dependencies (PyTorch, basicsr, facexlib, gfpgan), which can be challenging for users without experience in Python environments.

In practice

Price: Free
Commercial use: Commercial use: yes
Available on: cli, windows, linux
Licence: BSD-3-Clause
Runs locally: Yes, 4 GB VRAM recommended
FR interface: No

Security and data

Rating: Security: nothing to flag
Your data: Data stays local
Maintenance: Inactive for over six months
Last release: 20 September 2022

Links and repository checked on 25 October 2026 · added on 1 July 2021

How to install Real-ESRGAN

1. Install Python 3.7 or higher and PyTorch from pytorch.org. 2. Clone the repository: git clone https://github.com/xinntao/Real-ESRGAN.git then enter the folder: cd Real-ESRGAN. 3. Install dependencies: pip install basicsr facexlib gfpgan && pip install -r requirements.txt && python setup.py develop. 4. Download the desired model, for example RealESRGAN_x4plus.pth, into the weights/ folder. 5. Run inference: python inference_realesrgan.py -n RealESRGAN_x4plus -i inputs. 6. Security note: all processing remains local, no files are sent to an external server.

pip install basicsr facexlib gfpgan && pip install -r requirements.txt && python setup.py develop

In the same category

Whole category Video →