Implemented as a notebook, using tensorflow, I built a convolutional neural network able to categorize a movie by its poster with circa 65% accuracy.
You can choose whether to download and use the full dataset or use a smaller made-by-me subset, from the first variable of the notebook.
Note that a Kaggle account and an API key token are needed for the full dataset. Check how to get an API key token from here: Kaggle API authentication
Otherwise, the small testing subset is freely available from Hugging Face.