Translated text in an image runs longer: check fit before you publish

Translated text in an image is often longer than the original, so check each language for fit before you publish. A headline that fits in English can wrap, shrink or crowd a logo when the new words need more room. An edit model decides alone how to handle that, and the Sume API gives you no layout control beyond the prompt.

Where it goes wrong

• The new line is wider than the box, so the model shrinks it and it no longer matches the other lines.
• A button label grows past the button's edges.
• A two-line title becomes three and pushes into the photo.
• Numbers and units move, or a word wraps in the middle.

A review table you can copy

Make one row per language and one column per check. It takes minutes and keeps the review honest.

Who should review

Ask a fluent reader for each language, not a translation tool. The model can produce fluent-looking text that has a wrong word, and only a reader will see it. Keep the review to what the image shows: spelling, meaning, fit, and that nothing was added.

Languages that read right to left, or that do not use spaces, need extra care in the layout. This post does not cover them; check the vendor and Sume pages for each before you commit to a batch.

What to put in the prompt

Ideogram's launch post (read 2026-10-05) calls 4.5 the most precise edit model, and Morphic's partner page says it translates text in an image. Neither promises that a longer translation will fit your layout. So say what you want:

• "Keep every line on the same number of lines as the original."
• "Keep the same font size; shorten the wording if needed."
• "Keep the button text inside the button."
• Give the exact target text, quoted, when you have a translator's version, instead of asking the model to translate.

A fit routine

Run one language at a time from the original image, with one billed edit each. A completed edit is billed whether or not the text fits, per the Image API docs; failed and cancelled ones are not. Check each result against the original at the same zoom. For any line that overflows, choose between three fixes. Shorten the text in the prompt and re-run. Re-run at a higher quality tier. Or leave the line out of the image and put it on the page as live text.

For logos, prices and legal lines, the safe option is to draw the text in code or in your design tool and use the model only for the picture. That keeps wording exact, and it makes the next language a data change, not a new image.

If a translation does not fit, you have three honest choices. Shorten the translated wording with a person who reads the language, ask for a smaller type size in the prompt, or give the headline more room by editing the layout in a design tool. Do not accept a result that crops or overlaps text just because the words are right. Readers see the layout first and the words second. Keep the scope of this advice in view. It rests on the Sume docs and the vendor pages named in the sources, read on 2026-10-05, and on nothing measured by Sume. Where a behavior depends on your own images, such as how a model redraws a certain typeface, run a small pilot at the low quality tier and judge the result yourself before you plan a batch. Write down the prompt, the model id and the quality tier you used, so the run can be repeated. When the catalog or the docs change, re-read them; the live catalog is the contract, and a post is only a snapshot of it.