ESRGAN & SFTGAN
Installing this open source software can be a little tricky to newcomers. I'll try to walk you through the process.
I've compiled some useful techniques from various sources that will help you get more out these applications.
ESRGAN is an image upscaler that uses artificial intelligence to interpolate intermediate pixel values. The algorithm relies on pre-trained models to perform its upscaling. Generally, the closer the training data set used to train the model is to your desired input and output, the better the model will perform for your use case.
Technically, anyone with a large set of example image sets and a large amount of computational power and time could potentially train their own model to better upscale retro video game graphics. However, for the sake of these guides we will only use others' pre-trained models, which by default will upscale the input image by a factor of 4.
An interesting feature of ESRGAN is that it allows you to effectively blend the results of two models into one result; they call this "network interpolation" and it can be quite useful.
SFTGAN on the other hand is an image enhancer that uses Spacial Feature Transformation (SFT) to restore lost details of known feature types (brick, fur, grass, etc.). The important thing to remember here from a user stand-point, is that ESRGAN will increase the resolution while SFTGAN will enhance lost details.