Having two gans is not a problem. Although for different purpose it exists since a long time (cyclegan). What's important is that your loss and/or other constraints must be precise to make sure to avoid one gan learns both and other learns identity. It's also double the concerns for instability during training. Good luck!