Most captions fail for one reason
They were judged at full size on a good screen, and then everybody saw them somewhere else.
The photo goes into a feed as a thumbnail, appears in a search result at a couple of hundred pixels wide, or gets opened on a phone with the brightness down on a train. Text that looked considered at full size is a grey smudge in all three.
The fix is unglamorous: make it bigger than looks right, and give it enough contrast that it does not depend on the photo behind it.
Do not trust the picture underneath
White text over a sky works until the sky is bright. Dark text over a shadow works until somebody views it on a phone that has decided to brighten shadows for them.
Photographs are not uniform, and the part of the image behind your caption is the part you have least control over. If the words must be read, put something solid behind them: a filled panel, a semi-transparent strip, or a heavy outline around the letters.
It looks less elegant in the editor. It is the difference between a caption and a decoration.
Keep it away from the edges
Every platform crops. They crop differently on desktop and on mobile, they crop differently again for a preview card, and they change the rules without telling anybody.
Anything near an edge or a corner is the first thing to go. A date in the bottom right, a name along the top, a logo tucked into a corner: all of those are exactly where the crop lands.
If the text has to survive wherever the image ends up, keep it well inside the middle of the frame and accept the composition compromise. There is no way to negotiate with an automatic crop.
Type is drawn into the pixels
When you export, the words become part of the image. They are no longer text: they are a picture of text.
That has one advantage worth knowing. Nobody can select it, copy it, or extract it from the file, and it will look identical on every device because there is no font to be missing. A date stamp or a label burnt into a photograph is genuinely permanent.
It also means you cannot fix a typo without starting again from the original. Keep the original.
What people are usually doing here
Labelling a photo of a job so the person receiving it knows which unit, which floor or which date it refers to. Putting a name on a photograph before sending it to somebody who will have forty others. Adding a caption before a photo goes onto a page. Marking a screenshot with what it is showing.
All of those have the same requirement: it needs to be legible to somebody who is not looking carefully. None of them need an account, an upload or a subscription, which is most of what the alternatives ask for.
Nothing leaves your device
The photo is decoded in your browser, edited there and saved from there. No upload, no queue, no account and no cap on file size beyond what your own machine can handle.
Which matters more than it sounds for this particular job, because a photograph you are adding a name or a date to is quite often a photograph of somebody's property, somebody's work or somebody's family.
Common questions
How do I keep text readable over a busy photo?
Give it a background rather than trusting the photo. White text on a light sky, or dark text over a shadow, disappears the moment the image is resized or shown on a dim screen. A solid or semi-transparent panel behind the words, or a heavy outline, works on any photo. Relying on the picture underneath being the right brightness is the most common reason captions become unreadable.
What size should the text be?
Larger than looks right on your screen. Whatever you are making this for, somebody will see it as a thumbnail in a feed, in a search result, or on a phone held at arm's length. A caption sized to look elegant on a desktop monitor is unreadable at a quarter of the size, and a quarter of the size is where most people will meet it.
Will the text stay where I put it when the platform crops the image?
Not necessarily, and this catches people out constantly. Social platforms crop images differently on desktop and mobile and change their minds periodically. Text near an edge or a corner is the first thing to disappear. Keep anything that must survive well inside the middle of the frame.
Can I edit the text after saving?
Not from the exported file. The text is drawn into the pixels, so a saved JPEG or PNG is a picture of the words rather than the words themselves. Keep the original photo, and if you expect to revise a caption repeatedly it is worth writing down what you typed.
Is there a watermark on the download?
No. No badge, no name in the corner, no reduced resolution and no account to remove any of that. It runs on your device so there is nothing to recover the cost of.
Is the photo uploaded?
No, and on this tool it is worth more than it sounds. People add text to screenshots of messages, photographs of documents and pictures of their own children, which are exactly the files nobody should be handing to a stranger to process. It is opened, edited and saved by your own browser, so there is no size cap beyond what your device holds, no queue, and it works with the network switched off. The copy you save also carries none of the original camera metadata, because it is redrawn, not edited in place.