Is it possible to remove watermarks with a neuron?

Question

Watermark

Good afternoon! There are videos with watermarks that appear at regular intervals and in different places on the screen, with the same characters (letters + numbers), in order not to deal with manual blurring, I came up with the idea to convert frames into pictures, remove these marks on them and then assemble it back into a video.

Thus, there is a certain sequence of pictures where we can use the previous frame without a sign, a frame with a sign, and subsequent ones.

There is an idea to somehow train the neuron so that it paints over these watermarks using frames without them, or train it on a separately cut out this sign. Ideally, in order to understand where this sign is and take information from the frame where it is not, in this case, editing would not be noticeable at all.

I have basic knowledge of python.

Can someone tell me if there is already a similar trained neuron, or in which direction to dig, which libraries to use, of course, preference towards a ready-made solution, because it will take a long time to write and understand it yourself.
Thank you.

Answer 1

Basically you need something like this https://www.nvidia.com/research/inpainting/ .