Img translator / Photo translator
Free · no sign-up · runs in your browser
A free photo translator that keeps your layout.
Most ways of translating a picture give you back a different picture. This photo translator gives you back the one you uploaded, with the words in another language: OCR marks where every line of text sits, only the pixels of those letters are erased and repainted, and the translation is typeset into the same box at the same place. The artwork, the spacing and the brand colours are never touched, because nothing outside the letters is.
How this photo translator works
There are four steps, and only the third one asks anything of you.
- Load the photo. PNG, JPG or WebP. It is decoded into a canvas inside your own browser tab; the full-size file is never uploaded.
- Detection runs on its own. A downscaled copy goes out once to an OCR service, which reports where each line of text sits and what it says. Boxes appear on the preview. If one is wrong you can fix its text, move it, resize it, or drag a new box anywhere on the image by hand.
- You supply the translation. Pick your target languages and press Translate with ChatGPT. That copies a table already filled with your photo's own text and opens ChatGPT in a new tab. Paste, send, copy the reply, paste it back — the grid fills itself and every cell stays editable. If the strings already exist in a spreadsheet, upload a CSV, Excel or JSON file instead and skip the model entirely.
- Export. Each language is re-typeset in place and written out as a PNG, or all of them together as a ZIP.
What makes it a photo translator rather than a picture generator is the second half of step four. The original pixels are not regenerated from a description of them. They are still the pixels you uploaded, minus the letter strokes, plus new letters.
What a photo translator should not do to your image
Ask an image model to translate a picture and it makes a new picture. The font drifts. The logo comes back subtly wrong. A face changes. Fine print turns into fine-print-shaped texture. What you get is a convincing lookalike of the file you uploaded, which is fine for a mood board and unusable on a product page, an ad account or anything with a legal line in it.
This photo translator works the opposite way round. Erasing is done at the level of the letter strokes, not the bounding box: the pixels that belong to the old text are found, and each one is repainted by looking outward in twelve directions for the nearest known background on either side and interpolating between the two ends that agree. A stroke is only a few pixels wide, so the colour it gets comes from the background immediately beside it. Gradients keep gradating, diagonal stripes stay continuous, an illustration that ran behind the headline survives intact.
Then the new line is fitted. If the translation is longer — and it usually is, going from English into German or French — the type size drops, then the line wraps (per word in Latin scripts, per character in Chinese, Japanese and Korean), then it drops further, and finally the export is flagged rather than allowed to overflow. Colour, weight and alignment are read off the original line, so the result looks like the design it came from.
Where a photo translator earns its keep
Anywhere the picture is finished and only the words need to change. A photo translator that keeps your layout is worth more than a redraw in exactly those situations:
Product photos and marketplace listings
One listing image, five storefronts. The badge, the spec table and the price flag are re-typeset language by language while the product shot itself stays byte-for-byte the product shot. There is a longer write-up of this case on the e-commerce image translation page.
Ad banners and social creatives
A banner that already went through design review ships in eight locales without going back into the design file. Headline, subhead and button label change; the composition does not, so nothing has to be approved twice.
App screenshots and documentation figures
Store screenshots and help-centre figures get localised without rebuilding the screen in every language. The label pixels are replaced and the chrome around them is left alone.
Menus, signs, packaging and paperwork
Photograph a menu, a shelf label or a carton and read it back with the layout intact, so a price still sits beside the dish it belongs to and a warning still sits under the icon it belongs to. Position is meaning in these documents, and a flat wall of translated sentences throws it away.
Slides, charts and diagrams
Figures exported as images keep their callouts in place. Step numbers, arrows and captions stay where the diagram put them, which is the entire reason a diagram works.
How it differs from a camera translate app
Phone camera translation is built for reading. You point it at a sign, an overlay appears, you understand the sign, you move on. It is very good at that and this is not a replacement for it.
This photo translator is built for shipping. The difference shows up in three places. You get a file back — a real PNG at the original resolution, not a live overlay that disappears when the camera moves. You get editorial control: the detected text is editable before anything is translated, the translation itself is a table you can correct cell by cell, and approved marketing copy can be uploaded instead of generated. And you get every language in one pass — select twenty, export twenty PNGs and a ZIP, or run up to twenty images in a batch where repeated text across the whole set is translated exactly once.
What leaves your machine
One thing, once per image: a downscaled copy sent to the OCR service so it can find where the text sits. It is not stored. The full-resolution original stays in the tab. The exports are assembled in the browser and handed straight to your downloads folder. And no text — not the text found in your photo, not the translation — is sent anywhere at all, because this site calls no translation API and holds no model key. That is the reason the ChatGPT step is a copy-and-paste rather than a button that does it for you: the round trip is between you and a service you already chose.
You do not have to take that on trust. Open DevTools, watch the Network tab, and use the tool — loading an image produces exactly one request, translating produces none, exporting produces none. The private image translator page walks through the check and lists what the two anonymous page-view pings do and do not see.
The honest limits
Backgrounds that are solid, gradient, striped or lightly textured come out clean. Text sitting on a busy photograph — a headline across a crowd, a caption over foliage — is the case that still shows erase marks, because when the letter colour and the background colour cannot be told apart there is nothing to separate. That case is out of scope for now, and the photo translator says so on the export rather than quietly producing a smear.
The other limit is the ChatGPT hand-off itself. It cannot submit for you or fetch the answer back, so there is a copy, a paste, and a paste. In exchange, your words never touch this server, there is no account, and there is nothing to pay.
Questions
Is this photo translator free?
Yes — free, with no account and no watermark. Load an image, translate it, and export every language as PNG or as a single ZIP. There is no per-image charge. The only ceiling is a rate limit on the text-detection step, so that one visitor cannot exhaust it for everybody else.
How do I translate the text in a photo?
Drop the photo on the editor. Detection runs by itself and marks every line it finds. Pick the languages you want, press Translate with ChatGPT — that copies a table already filled with your photo's text and opens ChatGPT — then paste the reply back. The grid fills itself, and Export writes one PNG per language.
How many languages can I translate a photo into?
Twenty target languages, and you can select as many of them as you like in one run. Each selected language is exported as its own PNG, and all of them together as one ZIP. The interface itself is available in nine languages.
Does the photo translator work on a phone?
Yes. It is a web page, not an app, so it runs in Safari on iOS and Chrome on Android with nothing to install. Detection, editing and export all happen in the mobile browser; on a narrow screen the editor stacks the controls above the preview.
Which file formats does it accept?
PNG, JPG and WebP go in. PNG comes out — one file per language, or a ZIP when you export several at once. If your translations already exist somewhere, they can be brought in as CSV, Excel or JSON instead of going through ChatGPT.
What happens when the translation is much longer than the original?
The line keeps its original size if it fits. Otherwise it shrinks to 70%, then wraps — per word for Latin scripts, per character for CJK — then shrinks further, down to 35%. If it still will not fit, the export is flagged “text squeezed” rather than quietly clipped.
Do my photos get uploaded?
One downscaled copy of each image is sent to the OCR service to locate the text, and it is not stored. The full-resolution original is never uploaded, the exports are never uploaded, and no text — original or translated — is sent anywhere by this site. Separately from your file, the site counts anonymous page views; nothing about your image or your words is in that.