How does an LSTM network work?

Question

Neural networks

I am studying LSTM networks for my project. I studied and repeated this project .
Right now I'm trying to set up a network to handle streaming video. In the specified project, the network creates a description for the image, and they have 2 inputs - the image itself and the generated sequence.
What inputs should I have? After all, my network does not generate a sequence, but only analyzes it.
If you want to understand my project, then I have a document describing it in detail.
I would be glad if you could give advice on my project or send me a project or tutorial where the LSTM network processes video.