Downloads
Setup
How to use
Documentation
Examples
Subtitles

About

This is a simple guide to installing and using Real-ESRGAN with a few code enhancements that provide speed, fixes and quality of life improvements.

Real-ESRGAN is an Image and Video upscaling tool that improves the image quality.

Installation Guide 📖

Youtube Tutorial

Downloads

Download and install Anaconda

Clone/download Real-ESRGAN

Make sure to download Full Version of ffmpeg

(Optional): Add ffmpeg to Path Environment, Useful for manipulating video files

Setup

1. Open an anaconda prompt and navigate to where you downloaded Real-ESRGAN:

   cd {path to your real-esrgan repo}

2. Then create and activate the environment:

   conda create -n real python=3.11.7
   conda activate real

3. Now we need to set up some dependencies: (Important to follow the order)

   pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu121
   pip install chardet basicsr ffmpeg-python
   conda install ffmpeg
   pip install -r requirements.txt
   python setup.py develop

4. After that go to where you installed Anaconda:
   

   {path to Anaconda3}\envs\real\Lib\site-packages\basicsr\data\

   Open the file named degradations.py with a text editor then find and replace the following line8:

   torchvision.transforms.functional_tensor => torchvision.transforms.functional

5. Navigate to where you downloaded ffmpeg and copy everything from the full ffmpeg bin directory:

   {path to ffmpeg}\bin\

   Paste and Replace ffmpeg files in the pkg files here:

   {path to Anaconda3}\pkgs\ffmpeg-6.1.1-gpl_h66c0b5b_108\Library\bin\

6. Finally, Add func.py and inference_realesrgan_video_fast.py files from this repo to your Real_ESRGAN directory
   
   inference_realesrgan_video_fast.py is a modified version of the original file inference_realesrgan_video.py with code enhancements to speed up video processing speed.
   func.py is a little utility I made to make the process easier and more simple.
   
   Optionally you can also add the snippets.txt which contains some useful code snippets that you can copy and paste into Anaconda prompt to quickly run the process.

How to use ⚡

Inside your Real-ESRGAN folder, Find the inputs folder, Place the files you want to upscale inside

Copy and paste the following lines in an anaconda prompt to start processing all the files in the inputs directory quickly

cd {path to your real-esrgan folder}
conda activate real
python func.py

If you have your repository stored on a drive other than the one windows is installed on, Then remember to add a line to the start to boot into that drive for example:

d:
cd {path to your real-esrgan folder}
conda activate real
python func.py

If you want to process specific files or apply specific settings for each file then you can add lines in a note pad with customization for each file like the example below, then copy and paste it into anaconda:

python func.py -i inputs/myvideofile1.mp4 -n RealESRGAN_x4plus
python func.py -i inputs/myvideofile2.mp4 -o result_videos
python func.py -i inputs/myvideofile3.mp4 --suffix 4k
python func.py -i inputs/myvideofile4.mp4 -s 4
python func.py -i inputs/myvideofile5.mp4

func.py is just a helper tool to make the process easier and more accessible, If you wish for more control then you can make lines that call the inference files directly:

# calling the inference_realesrgan_video_fast directly
python inference_realesrgan_video_fast.py -n realesr-animevideov3 -i inputs/myvideo.mp4 -o results -s 2 --link --suffix fast_4k

# calling the original inference_realesrgan_video directly
python inference_realesrgan_video.py -n realesr-animevideov3 -i inputs/myvideo.mp4 -o results -s 3 --suffix default_4k

Documentation

func.py has a few arguments that allow you to customize the behaviour

Usage: python func.py [options]...

    -h --help                   show this help message and exit

    -i --input                  Specify the input file/directory, Default: inputs
    
    -o --output                 Specify the output directory, Default: results
    
    -s --outscale               The final upsampling scale of the image, Default: 2

    -n --model_name             Model names: realesr-animevideov3 | RealESRGAN_x4plus_anime_6B | RealESRGAN_x4plus |
                                RealESRNet_x4plus | RealESRGAN_x2plus | realesr-general-x4v3Default:realesr-animevideov3
    
    -sf --suffix                Suffix added to end of the output file, Default: 4k
    
    -dm --dont_move             Stops the file from moving to the done folder, Useful if you are planning to process the same
                                file multiple times, or if you just dont like this behaviour
    
    -da --disable_audio         Do not include audio stream in result
    
    -ds --disable_subtitles     Do not include subtitle stream in result

Examples

# specify custom input or output 
python func.py -i A:/real/inputs/video              # folder
python func.py -i inputs/myVideo.mp4                # file (partial paths will search inside the project directory)
python func.py -i inputs/myVideo.mp4 -o my_results  # file with output


# specify custom scale
python func.py -s 4

# specify custom model
python func.py -n RealESRNet_x4plus

# specify custom suffix
python func.py -sf 4k_result

# stop file moving location after process is done
python func.py -dm

# do not include audio track in result
python func.py -da

# do not include subtitle track in result
python func.py -ds

Subtitles

The code for Subtitles is quite experimental and will likely break, it's honestly best to disable it -ds and find another way to extract the subtitles from the original work and add it to the 4k result