Guide

How to Translate Product Images Without Changing the Design

"How do I translate the text in this product image?" has been solved for years — point Google Lens at it and you can read what it says. That's not the question that actually matters if you're a seller. The real question is: can you take the output and upload it as a new listing image today, without a designer touching it first? That's a much higher bar, and most tools that claim to "translate images" were never built to clear it.

Quick answer: a listing-ready translated image needs to pass five specific checks — original size/background, clean text erasure, matched font weight/color/outline/shadow, sensible handling of longer translated text, and a usable file-naming convention for multi-language exports. Below is how to test any tool against each one, plus an honest look at where Google Lens, Canva, ImageTranslate and KeepFormat each actually fit.

"Reading the text" and "replacing the listing image" are different problems

Google Lens and Google Translate's image feature are genuinely good at the first problem. They'll tell you what a Chinese product image says in seconds. But by design, they overlay a plain translation on top of the image or your camera view — they don't reconstruct the original typography, and there's no batch export for a catalog of listing images. That's not a flaw; it's just a different product built for a different job.

Translating a product image without changing the design means something more specific: the output has to survive being dropped straight into an Amazon, Shopify, Shopee, or Etsy listing, next to your other (untranslated) images, without looking out of place. That's the bar the five checks below are testing for.

5 things to check before you trust a translated product image

1

Original size and background

Check: Does the output match the source image's exact pixel dimensions, and is the background — color, gradient, or product photo — untouched?

Marketplaces have fixed image-size requirements (Amazon's main image rules, Shopify's theme grid, Etsy's listing dimensions). A translated image that comes back a different size or aspect ratio doesn't just look different — it may not fit the slot at all.

2

Clean text erasure

Check: Where the original text was, is there a seamless continuation of the background — or a visible solid-color box, blur patch, or white rectangle?

This is the single most common tell that an image was machine-translated. A flat color patch sitting over a gradient or a product photo is the difference between “this looks professionally localized” and “this was obviously edited.”

3

Font weight, color, outline, and shadow

Check: Does the new text match the original's boldness, exact color, any outline/stroke, and any drop shadow — or does it default to a plain system font?

E-commerce marketing text is rarely plain — bold headlines with an outline or gradient fill are the norm on banners and badges. A tool that only handles plain text quietly fails on exactly the text that matters most (the price callout, the “NEW” badge, the headline).

4

Long text: auto-scale or wrap

Check: Chinese-to-English text often expands 30–50% in character count. Does the translated text shrink to fit, wrap onto a second line inside the same box, or overflow outside it?

This is where a lot of tools that work fine on short labels break down on longer marketing copy. Overflowing text runs into other design elements; text that's shrunk too aggressively becomes illegible on a mobile thumbnail.

5

Multi-language batch export and file naming

Check: If you're translating one image into five languages for five marketplaces, do the exported files come out with a naming convention you can actually manage — or do you get five identically-named downloads?

This is a workflow problem, not a translation-quality one, and it's easy to overlook until you're managing 40 SKUs across 5 storefronts. A tool can translate perfectly and still cost you an afternoon of manually renaming files.

A file-naming convention for multi-language exports

Whether or not the tool you use has automated batch export, this convention keeps a multi-marketplace, multi-language image set sane:

{original-filename}-{language-code}.{ext}
→ hero-banner-en.jpg
→ hero-banner-de.jpg
→ hero-banner-ja.jpg

Keeping the original filename intact means you can always trace a translated file back to its source — useful the moment you update the source image and need to know which translations are now out of date.

How to test any image translation tool yourself

  1. Pick one real listing image with at least one headline and one price/badge callout.
  2. Run it through the tool and open the result at full size, not a thumbnail.
  3. Zoom into where the original text was — look specifically for a flat color patch.
  4. Compare the new text's weight, color, and any outline against the original.
  5. If the translated phrase is noticeably longer than the source, check whether it shrank, wrapped, or overflowed.
  6. If you need multiple languages, note exactly how the output files are named.

Google Lens vs. Canva vs. ImageTranslate vs. KeepFormat

We haven't run controlled, side-by-side tests against the other three tools here, so rather than invent a score, here's what each one actually is and claims — with sources — so you can run the checklist above yourself.

Google Lens / Google Translate

Built for understanding text you're looking at, not producing a new file. Overlays plain translated text on the live camera view, or over a photo, without matching the original typography.

Not built for this job. There's no font/color matching and no batch export — it answers “what does this say?”, not “give me a file I can upload.”

Source →

Canva (Image Translator app)

Uses OCR to detect text in an uploaded image, translates it, and reconstructs the image with the translated text — Canva's own description is that it applies “digital image processing” to produce the new picture.

A reasonable general-purpose option if you're already inside Canva, with a free-tier cap of 50 pages. Since it's a general design tool rather than an e-commerce-specific one, test items 2–4 above directly on your own banner before relying on it for a listing image.

Source →

ImageTranslate

A dedicated product-image translation tool that explicitly targets Amazon sellers and dropshippers. Their own site describes “original text erasure” that preserves the surroundings, plus font, color, and size prediction, with batch conversion across 40 languages.

The closest like-for-like comparison on paper — same target use case, same claimed dimensions. We haven't run controlled tests against it, so test your own image on both and compare items 2 and 3 directly.

Source →

KeepFormat

Translates the whole image in one pass, aiming to keep the exact background, style, and layout while replacing only the text. Built on the same principle as our PDF tool: the file should still look like the original, not just be readable.

Currently one image at a time, one target language per job — if you need five languages, that's five runs today rather than a single batch export (see the naming convention below in the meantime). We'd rather say that plainly than have you find out after signing up.

Try it →

Frequently asked questions

Can Google Lens translate a product image for me to use as a new listing photo?

Not really — Google Lens and Google Translate's image feature are built to help you understand text you're looking at. They overlay a plain translation on the image or camera view rather than reconstructing the original design, so the result isn't something you'd upload as a finished listing image.

Why does my translated product image have a plain box where the Chinese text used to be?

That happens when the tool erases the original text with a flat color fill instead of reconstructing the background underneath it — it's the most common visible sign of a machine-translated image, especially on gradients or photographic backgrounds.

How should I name files when translating one product image into multiple languages?

A simple, sortable convention works best: {original-filename}-{language-code}.{ext}, for example hero-banner-en.jpg, hero-banner-de.jpg, hero-banner-ja.jpg. Keep the source filename intact so you can always trace a translated file back to its original.

What happens if the translated text is longer than the original?

It depends entirely on the tool. Some shrink the font to fit the original text box, some wrap it onto an extra line within the same space, and some let it overflow outside its original boundary. Since Chinese-to-English text commonly expands 30–50%, this is worth checking specifically on any marketing copy longer than a short label.