I planning to fine-tune real-ESRGAN for my project and as I do my research in its github repo there are 2 cases for fine-tuning the model:

1. Generate degraded images on the fly (meaning I only need to provide the high res image to the model and it will automatically generate the low quality image by its self through blurring, adding noise, compressing,etc)
2. Use my own paired data

As for the first case, I wonder if I can add more degradation effects, for example changing the color to greyscale or vintage yellow to make it look like old picture?