These are high-quality upscale models based on an exceptionally strong but still lightweight architecture - SAFMN
The architecture file was modified by me to prevent the artifacts that are present in the default architecture from the original paper.
To use these models with Chainner you must replace the default SAFMN architecture file with the attached one (safmn.py). Assuming you're using a portable Chainner installation, the directory of the SAFMN architecture is here: C:\chaiNNer\python\python\Lib\site-packages\spandrel\architectures\SAFMN__arch
2x_SAFMN_PureScale - a model for high-fidelity upscaling, handles JPEG compression
2x_SAFMN_PureScale_sharper - a model for high-quality upscaling with a deblurring component, handles JPEG compression
4x_SAFMN_PureScale - the ultimate model for high-fidelity upscaling; handles JPEG compression as well