How to optimize a neural network bot for multiple users?

Question

CUDA

I am writing my first bot (on aiogram), in which I have embedded a style-transfer neural network on pytorch. My problem is this: the neural network works for some time (40-50 seconds), and then sends a response. If a user writes to the bot at this time, he, of course, will ignore it. Advise how best to organize the reception of photos from users (maybe their id should be added to the queue and retrieved from there, or something else). Thank you.