Always add a stroke over a photograph
Text drawn onto an image sits on whatever happens to be underneath it. A light caption over a bright sky disappears entirely; a dark one over a shadow does the same.
A contrasting outline behind the text fixes this regardless of what is below, which is why it is the setting worth reaching for first. Flat-coloured text works on flat backgrounds and nowhere else, and a photograph is by definition not flat.
The [msm_tool slug="add-text-to-image"] offers the stroke alongside the fill, and on any real photograph you want both.
Size as a proportion, not in pixels
The size is set as a percentage of the image rather than an absolute value, so a caption occupies the same visual proportion whether the file is a thumbnail or a full-resolution photograph.
That matters because the same image is almost never used at one size. A caption specified in pixels is correct at exactly one resolution and wrong everywhere else — enormous on a thumbnail, invisible on the original.
Leave room at the edges
Social platforms and preview cards crop images without asking, and they crop from the edges inward. Text near a border is the first thing to be cut off.
The habit worth forming is to keep text within the middle eighty percent of the frame. It costs a little composition and it means the caption survives being reposted, shared and re-cropped by systems you do not control.
This is not a substitute for real text
Words drawn into an image cannot be read by a screen reader, cannot be selected or copied, cannot be translated by a browser, and are not indexed by search engines.
So the rule is straightforward: if the text carries meaning, put it in the page as actual text and use the image behind it. Drawing it into the picture is for cases where the image travels on its own — a social card, a shared graphic, a printed asset — and even then, the meaning should exist somewhere else too.
Where an image with text does end up on a page, the alt text should include the words in the image. That is not optional; it is the only way anyone using assistive technology gets the content at all.
Opacity is separate from colour
Keeping them separate lets a caption sit lightly over an image rather than dominating it. A watermark-style credit at thirty percent reads as attribution; the same text at full opacity reads as a headline.
Judge it on the photograph rather than on a preview panel — a busy image swallows faint text in a way a flat background does not.
Choosing an export format
PNG for text over graphics or flat colour, where sharp edges matter and JPEG artefacts cluster around high-contrast boundaries — which is exactly what letterforms are.
JPEG is acceptable over photographs at a reasonable quality, and it softens fine lettering slightly. For small text or thin typefaces, PNG is the safer choice even at the cost of file size.
It cannot be moved afterwards
The text is drawn into the pixels. Repositioning means starting again from the original, so keep it — particularly for anything you may need to reproduce in a different size or language later.