Are there generative neural networks that use "forbidding" datasets?

Question

Neural networks

Are there generative neural networks where, in addition to the data that
needs to be generated, the input is also data that cannot be generated like?
If it exists, what is it called?
I formulated it a little crookedly, but I think the essence is clear
. After all, according to the logic, such a network ("generate like this, but don't generate like this") should be more flexible and accurate than just ("generate like this")
In my opinion, such a network can be trained on smaller datasets.
After all, any training is based on "can" and "shouldn't", and not just on "can".
From an early age, a child is taught that one should not stick one's fingers into a socket, play with fire, and so on. They don’t tell him: “You can do everything from this list,

Answer 1

1 link?