Something that quietly eats time on our team: the visuals in internal runbooks and onboarding docs. The text is usually fine, but the pictures are either screenshots full of hostnames and IPs, or photos of a whiteboard taken at a bad angle that nobody can read six months later.
What has worked reasonably well lately is keeping two separate pipelines. For anything that shows a real environment, screenshots only, always sanitised, never regenerated. For concept drawings, though, like how a backup chain flows or how a restore point timeline is supposed to look, I sketch it on paper, photograph it, and run it through an AI image tool to get a clean vector-ish illustration. I have been using Muse Image (https://muse-image.co/) for this because it takes several reference images at once, so I can feed the sketch plus two screenshots of our actual UI and keep the layout close to the original.
The one rule I would insist on: never let the model touch a real screenshot. It will happily invent menu items and version numbers, and then someone follows the doc and gets lost. Concept diagrams only, and label them as illustrative in the caption.
Curious how other people handle this. Do you maintain a diagram library, redraw everything by hand, or just accept the ugly screenshots? And has anyone found a way to keep the drawing in sync when the product UI changes every release?