In the context of publishing, editors want original figures because previously used ones are copy-righted.
As far as self-plagiarism is concerned: First, this is a concept that covers two different fields: (1) Student work submitted for evaluation and (2) academic articles. The prohibition of self-plagiarism for (1) is the impact on the evaluation of student work. For (2), it is about someone inflating their work (usually for self-promotion) by publishing the same idea multiple times. Using criteria from context (1) for context (2) is misleading (even though I will get down-voted for this opinion).
Therefore, witty formulations from the introduction copied from one work to another without marking them as such is self-plagiarism in context (1) and not in context (2). Using exactly the same idea to write to entirely different articles should be self-plagiarism in context (2) but is usually not in (1). Of course, there is a very large gray zone here.
In the context of your question: Figures can be very complicated and involve art work. I am thinking of a figure showing the different style of bronze age axes between two different sites. That figure should not be copied from one work to another without clearly marking it. Second, figures can be very utilitarian. Assume you have one data set that you use for both papers. You create a scatter graph using mathematica or matplotlip.pyplot. I do not think it is self-plagiarism if you end up using exactly the same figure. The creative novelty is in the data, not the depiction. Changing minor things such as colors or sizes does not change the original sufficiently, but it will make your editorial house happy because it is now harder to claim a copy-right violation.
Should you mark this clearly? The US patent law uses a fictitious "competent engineer" to test for the novelty of a potential invention. A similar standard should appear here, because otherwise terse academic writing will be filled with (adapted from) notes that are entirely superfluous. A figure is often just a way to describe data. If the same data is reused, that reuse needs to be flagged. But changing fonts on a simple scatter plot is silly.
To give a final example: You compare the population of Uruguay and Burundi over the last century from raw data. You cite the data source. The drawing of a graph is so simple that the drawing is not cite-worthy.