For a while every article on this site was pure prose. Not one picture, across thirteen pieces, several of which are about interface work and would obviously be better with one.
The reason was not neglect. It was that "this article needs an image" is a thought you have while reading, and by the time you are in a position to make an image you are somewhere else entirely, looking at a list of articles and no longer able to remember which paragraph wanted a picture and what of. The knowledge was being lost between noticing and acting.
So I did something that felt wrong when I typed it: I put empty frames into the articles, on the live site, where readers can see them.
The question that decides everything else
Before any of the design, one question has to be answered, and answering it wrongly makes every subsequent choice wrong too.
Who sees this?
If the answer is "only me, in development", you are building an authoring tool. It can say TODO, it can be magenta, it can be as ugly as it needs to be, and none of that costs anything because no reader will ever meet it.
If the answer is "everyone", you are publishing a piece of page. It has to be written for a reader, it has to look deliberate, and every convention from the authoring-tool version is now actively harmful.
I put the question to the person whose site it is, expecting the first answer, and got the second. His reasoning was better than mine: the place he reads his own articles is the live site. A marker visible only in a development build is a marker he will never encounter at the moment he could act on it, which is the entire problem it was meant to solve.
What that changes
Everything, and all in the same direction: away from the vocabulary of a to-do.
No dashed borders. A dashed outline is the universal sign for "not finished". It is the right signal in a design tool and the wrong one on a published page, where it reads as broken rather than as intentional. The frames use exactly the border, radius and shadow a real photograph gets on this site. They are the frame, empty.
No "placeholder", anywhere a reader can see. The word describes the thing's status in my workflow, which is not a reader's concern.
A caption, not a note. Every frame carries a sentence saying what the picture would show, written the way a caption is written: "The parts laid out on the desk before anything went in, which is the last moment the build looks like its shopping list." That sentence is doing two jobs. For a reader it is a small piece of the article that happens not to have an image attached yet. For me it is a brief, and a much better one than "image here", because writing it forces the decision about what the picture is actually of.
Here is one, since an article arguing that empty frames can be presentable should be willing to show you an empty frame rather than describe one.
Nothing announced to a screen reader as an image. There is no image. The frame is decorative, the glyph inside it is hidden, and the caption is the whole of what gets read out. Announcing an empty box as a picture would be a worse lie than saying nothing.
The unexpected benefit
Writing the caption first turned out to be the useful part, and I did not see it coming.
An image slot with no caption is a wish. An image slot whose caption says "the same room twice: the golden hour the model reached for, and the ordinary Tuesday light it actually gets" is a shot list. It tells you what to photograph, from where, and why, months after you had the thought.
It also kills the bad ideas early. Several frames I started to write did not survive their own caption, because when I tried to say what the picture showed, the honest answer was "the thing the paragraph already said". That is decoration, and a sentence is a cheaper place to find that out than a photo shoot.
And then most of them went away
Here is the part I did not expect to be writing.
Eight frames went in. Four were interface diagrams, and those turned out to be things I could just build: real components, real numbers, laid out in the site's own design system, which is a better answer than a photograph would have been. Four were photographic, and of those, two were in articles that were later unpublished for having nothing to do with the subject of this site.
So the frames are gone, and the exercise still paid for itself twice. (Building those four turned out to have its own lesson about clever CSS, but that is a different article.) Once because "what would the picture be of" is the question that turns four of them into something buildable that afternoon. And once because holding a space makes you look at whether the space is worth holding, and for two articles the honest answer was that the article itself was not.
An empty frame is a promise to a reader. That turns out to be a useful discipline precisely because it is uncomfortable: you find out quickly which promises you actually intend to keep.
If you want to try it
Keep it to a handful. If every article has a picture-shaped hole, none of them is a place where a picture matters, and the site starts to read as unfinished rather than as considered.
Write the caption as though the image already exists. If you cannot, you have learned something.
And make the frame indistinguishable from a real one. If it looks like scaffolding, it is scaffolding, and scaffolding on a published page is just a mess with a plan behind it that nobody else can see.