Optimizing ML model issues

I'm trying to convert my model (https://github.com/rtolps/TensorflowCats2DogsUGATIT ) that converts cats into dogs using UGATIT into a format Lens Studio can use. It generates a frozen graph file (.pb) but it's absolutely huge (40MB). I've tried a couple things to trim it down but nothing quite works. Additionally, a couple nodes (minimum) in the frozen graph are not supported by snapchat. I'm also not even sure the input/outputs are even in the right colorspace or whether they would work or not per SnapML's requirements. How do I trim down the size to below 10MB and prune and fix the inputs and outputs, essentially, how on earth do I get this to actually work? I would be really grateful if Snapchat released more Collab Notebook examples of their fantastic models. I am primarily an artist and hardly a software engineer and I'm very much in the dark here.