You can't extract any more information from the image than what the satellite originally acquired - you could make an "error correcting" model that creates an image that appears to have higher fidelity, but the additional detail is just a representation of what the model is filling in based on the other images it has been trained on, rather than an actual increase in resolution.