You draw it first for the group meeting, where nobody minds that the axis labels are the defaults. Then for the draft, where your supervisor asks what the shaded region means. Then for the journal, which wants 300 dpi CMYK at a column width you have to look up. Then for the talk, where it turns out nothing under 18 pt can be read from the back of the room.

Four versions of one figure, and by the fourth nobody can remember which file made which.

Decide the claim before you open the software

Almost every redraw comes from the same place: the first version was made to show the data, and every version after it was made to make a point. Those are different drawings. A figure that shows the data answers “what did you measure”; a figure that makes a point answers “what should I conclude”. Write the one-sentence conclusion above the panel before you start, and the chart type, the axis range and what gets a colour all follow from it.

If you cannot write that sentence, the figure is not ready — and no amount of redrawing will fix it.

Draw once, at the largest size you will ever need

The poster is the largest. Work at that size, in a vector format, and export down. Scaling a vector down is free; scaling a 96 dpi screenshot up is a new afternoon. Keep the source file next to the data that made it, with the script that generated it if there was one.

A figure you cannot regenerate from the data is a figure you will redraw by hand every time a reviewer asks a question.

Set the type once, too

One typeface, three sizes: label, axis, annotation. Choose them once for the whole project and every later figure inherits them. Most of the visual mess in a thesis is six figures made in four programs on four different days.

None of this is about aesthetics. It is about the fourth redraw not happening.