What I learned routing prompts between 20+ AI image and video models

When we started building DREAMMMER, I assumed the hard part would be the UI. It wasn't. The hard part was deciding which model should handle which request. Some notes from the last few months, in no p