What is image translation?
Image translation is the process of finding the text inside an image file, translating it into another language, and putting the translated text back into the image in place of the original. The input might be a photo of a street sign, a screenshot of an app, a scanned contract, a social media banner or an animated GIF. The output you usually want is the same picture — same layout, same colors, same background — with the words in a new language.
That last part is what separates image translation from ordinary text translation. Text in an image isn’t text yet: there is nothing to select, copy or paste. So people fall into a familiar loop — retype the words out of the JPG, paste them into a translator, then rebuild the graphic by hand, once per language. An AI image translator removes that loop by chaining three steps together: optical character recognition (OCR) to read the text, AI translation to convert it, and a rebuild step that writes the translation back into the layout.
Image, picture or photo translator?
People search for all three, and for the job they describe they are the same thing: a file with text in it. What changes the result is the file itself — a sharp screenshot, an angled phone photo and a 300 DPI scan each behave differently. This guide covers all three, plus GIFs and image-heavy documents.
Image translation is one part of the wider world of AI translation. If you want the bigger picture — how AI translators work, how accurate they are and how to translate text, web pages and messages — start with our complete guide to AI translators, then come back here for everything specific to images.
Why image translation matters
Visual content is everywhere text used to be: product photos with labels, infographics, training diagrams, UI screenshots in help docs, ad creatives in a dozen sizes. Each one carries words that need to reach an audience in its own language. For one image, a free camera app is fine. For forty product images in six languages, the manual route turns into a week of rework — and different people translating different images means the same product name comes back three different ways across one catalog. The rest of this guide walks through how to avoid that, by image type, by language and by use case.
Photos vs. screenshots vs. scanned images
Every image goes through the same pipeline — recognize, translate, rebuild — but where the text comes from changes how well the first step works. Pick the tool that matches your input:
Photos taken on a phone
Photos are the hardest images to translate. Perspective skew, glare, shadows and low light all degrade character recognition before translation even starts, and the text sits in a scene — a package label, a whiteboard, a handwritten note — rather than on a clean page. The Photo Translator runs OCR on the photo, shows you the recognized text in the editor so you can fix any character the OCR pass misread, and gives you back the same photo with the translation in place. There’s no app to install: it runs in your phone’s browser, and you can upload a single shot or a whole folder at once. If you only need the words, you can export the extracted text instead of the image.
For the best result, hold the camera straight on, fill the frame with the text and avoid glare. A bad photo produces a bad translation: the translation step can’t repair characters that were misread upstream.
Screenshots and on-screen text
Screenshots are usually the cleanest input you can get: the text is sharp, level and high-contrast, which makes them ideal for AI translation. App interfaces, error messages, chat threads, slides and software documentation all fall here. Upload screenshots one at a time or in bulk, choose your target languages, review the result in a live preview and download them in the original format. Because UI terms repeat across dozens of screenshots, add button labels and feature names to a glossary so they translate identically in every file.
Medical technology company Smith+Nephew runs its in-house translation team on Smartcat, keeping translation memories and glossaries in one workspace; its case study reports a 4x faster turnaround compared with its previous agency SLA (read the Smith+Nephew case study).
Scanned documents
A scan is a picture of words, not words. Translating one is a two-stage process: OCR converts the image of the page into machine-readable text, including text inside tables and headers, and only then is that text translated. Smartcat runs both stages in one upload and preserves the page layout, so what you download still looks like the document you uploaded. Scanned document translation works on scanner files and on straight-on phone photos of pages; scans should be at least 300 DPI.
Some scripts get dedicated handling. Scanned Chinese, for example, lets you set Simplified or Traditional (down to the regional variant) as the source, and segments the recognized text into words before translating, because written Chinese has no spaces — one wrong split changes the meaning even when every character was read correctly.
If your scan is a multi-page PDF rather than an image file, use a PDF workflow instead, which keeps the multi-page structure — our PDF translation guide covers scanned PDFs, OCR parsing options and layout in depth, and the scanned PDF translator is the place to start.
How OCR turns an image into translatable text
Optical character recognition is the step that makes everything else possible. OCR scans the pixels of an image, finds the regions that contain text — headlines, labels, captions, text inside diagrams — and converts each one into characters a translation engine can work with. In an OCR image translation workflow, that extraction and the translation happen in one pass, so you never handle the raw text unless you want to.
What affects OCR quality
OCR accuracy falls with resolution and rises with contrast. Sharp, straight-on images translate reliably; blurry or heavily compressed ones lose characters. A few things make a noticeable difference:
- Resolution. Scan at 300 DPI or better. Scripts with dense characters, such as Chinese, are less forgiving than Latin text, because the difference between two characters can be a single short stroke.
- Angle and light. Keep the page or sign flat and evenly lit. Glare and shadows hide strokes.
- Script direction. Right-to-left scripts like Hebrew and Arabic need the text direction detected before lines are ordered, especially when a line mixes Hebrew or Arabic with an English name or a number.
- Handwriting. Handwriting is read only by the image-based OCR pipeline, with accuracy tracking input quality. Treat a handwriting result as a draft to check, not a translation to publish.
- Busy backgrounds. Text on photographic backgrounds or in stylized logotypes is detected but rarely re-rendered perfectly, so plan on a manual touch-up.
The Hebrew picture translator page is a good illustration of why script-aware OCR matters: Hebrew runs right to left, five letters change shape at the end of a word, everyday text omits vowel marks, and cursive handwriting shares almost no letter shapes with print.
Image to text, then translation
Sometimes you don’t need a translated picture at all — you need the words, in your language, to paste into an email or a report. Smartcat keeps the extracted source text visible and editable alongside the translation, so a bilingual colleague or reviewer can check what the OCR actually read. That is the step that catches visually similar characters misread on a low-quality scan, which otherwise produce fluent, confident, wrong output.
Translating OCR output is also where AI translation earns its keep over word-by-word tools: the engine sees whole sentences in context and applies your glossary, so a product name extracted from image 3 and image 40 comes back identical.
How image translation works in Smartcat
Whether you start from a single picture or a campaign’s worth of banners, the workflow is the same five steps — and every step happens in the browser.
Upload your images
Drag one or many images into a project. The image translator reads common formats including JPG/JPEG, PNG, TIFF, BMP and GIF via OCR, up to 30 MB per image, with no cap on how many files go into one project. You can also ask Smartcat AI to translate an image instead of creating the project by hand.
Pick your languages and image type
Choose one source and as many targets as you need, from 280+ languages. For marketing visuals, choose the “Marketing & Visuals” image type to use the generative pipeline; for scans and screenshots, “Documents & Scans” uses the OCR pipeline with manual layer editing.
AI translates and rebuilds
Smartcat detects every text region, removes it from the background, translates it using your glossary and translation memory, and re-inserts the translated text into the original layout as editable text blocks.
Review and adjust
In the Image Translation Editor, a preview of the translated image sits above the text segments. Drag text boxes into place, change font, size, color and alignment, or hide text such as a logo so it stays untranslated. For anything high-stakes, invite a colleague or a vetted reviewer from Smartcat’s Marketplace without leaving the project.
Download in the original format
Export the same image, same layout, new language. Batch jobs download together, so a set of images translated at once comes back as one set.
For a click-by-click walkthrough, including editor shortcuts and the current list of supported fonts, see the help article Translate text in images with the AI Image Translator.
OCR pipeline vs. generative pipeline
Smartcat offers two ways to rebuild a translated image, and choosing the right one at project creation saves rework later.
| Factor | Documents & Scans (OCR pipeline) | Marketing & Visuals (generative pipeline) |
|---|---|---|
| Best for | Scans, screenshots, document images | Banners, ad creatives, social graphics, packaging, infographics |
| How the image is rebuilt | Background removed, translated text re-inserted as editable text blocks | AI generates the final image in the original style |
| Manual layer editing (move, resize, font) | Yes | No — edit text in the segment grid |
| Glossary & translation memory | Yes | Yes |
| Long translations | Adjust line breaks and font size yourself | Soft warning when text runs long; regenerate if needed |
The generative option is documented in Translate marketing images with AI (Image Translation 2.0). Final image generation typically takes up to 30 seconds, and a Regenerate button creates a new version if the first output isn’t right.
Keeping design and branding intact
For marketing and product teams, the translation is only half the job. A translated banner that overflows its button, swaps the brand font for a default one or leaves ghosting where the original text was isn’t publishable. Design-safe image translation means the background, fonts, alignment and colors survive the trip.
Marketing images, banners and ads
Smartcat’s generative pipeline was built for exactly this: it extracts the text, translates it with your glossaries and translation memory, then generates the final image in the original brand style — no manual in-painting or font matching. The zero-touch image localization update describes the engine’s automatic font matching, layout preservation and text-expansion handling, plus segment-based editing so you control the wording before the image is generated. When Smartcat first launched its AI Image Translator, the goal was the same: translate an image into several languages at once without rebuilding the design for every market.
Text expansion is normal.
Many languages run longer than English, so a text box sized for the source can overflow. That’s why the OCR pipeline keeps every text box editable — shorten a line or drop the font size in the browser rather than reopening a design tool — and why the generative pipeline warns you when a translation is significantly longer than the original.
Translating inside Canva and Figma
If your visuals are still in the design tool, translate them there rather than exporting flat images. The Canva translator brings a Smartcat agent into Canva’s app panel; it translates text in designs, including on-image text in graphics, infographics and social posts, using your glossary and translation memory (read the Canva Agent announcement). For product and UI design, Smartcat’s Figma translation plugin translates designs into 280+ languages with live previews, and lets reviewers work without paid Figma seats; setup is covered in Connect Figma with Smartcat.
GIFs and animated images
The GIF translator reads the text on a GIF, translates it and puts it back in place. For animated files, the GIF translator page says every frame that carries text is translated and the file is rebuilt with its original timing and loop, with your glossary applied across all frames. Two related pages cover GIF translation with AI and translating GIFs to English.
Images on websites and in documents
Images are the part of a web page that ordinary page translation misses: the words are pixels, not HTML. Smartcat’s pages on translating website images and translating images on web pages explain why consumer tools struggle with embedded and dynamic web images, and the image translator Chrome extension translates text within images without leaving the page. For the rest of the site — HTML text, SEO metadata, CMS integrations — see our website translation guide.
Images embedded in Office files need one extra step: when you translate a .pptx, the slide text is translated but text inside the pictures isn’t. The help article Translating images embedded in PowerPoint files shows how to extract those images, translate them and re-insert them. Our document translation guide covers the rest of the file-translation workflow.
File formats and batch image translation
Most image translation searches name a format — “translate PNG”, “JPG translator” — but the format matters less than the content. The OCR pipeline reads JPG/JPEG/JFIF, PNG, TIF/TIFF, BMP, GIF, DCX/PCX, JP2/JPC, DJVU, JB2, image-only PDF and Adobe Illustrator (AI) files, and exports in the same format you uploaded. Each image can be up to 30 MB.
PNG, JPG and other image files
PNG is common for screenshots, UI graphics and anything with transparency; JPG is the default for photos. Both go through the same recognize-translate-rebuild pipeline. The format-specific pages below walk through each:
Translating many images at once
The difference between translating forty images and translating one image forty times is shared context. When you translate multiple images at once, every image in the batch uses one glossary, so brand names and recurring terms are translated identically in every file, and one translation memory, so a sentence repeated across twenty images is translated once and reused. One batch can target several languages at the same time, and the finished set downloads together.
Batch translation is also where image translation becomes a team workflow rather than a one-off task: a whole campaign’s banners or a catalog’s product images upload as one project, a reviewer checks them in one place, and their edits feed back into translation memory for the next batch.
Translate images by language
Image translation demand is heavily shaped by language. Most searches name a direction — Japanese to English from a photo, English to Gujarati on a poster, Arabic to English from a scan — and some scripts are much harder to recognize than others. Smartcat’s image tools pair any of 280+ supported languages with any other, and the pages below are language-specific starting points, grouped by script and region.
Translate images into English
Japanese, Chinese and Korean
CJK scripts are among the most-requested image translations, and among the hardest for OCR: thousands of characters, strokes that differ by a millimeter, and in Japanese, text that may run vertically. Smartcat’s vertical Japanese text translator handles traditional top-to-bottom tategaki in images and screenshots, and the kanji photo translator covers kanji in photos. Note that the scanned Chinese workflow handles horizontally set Chinese; vertical Chinese columns aren’t supported there.
Hindi and other Indian languages
Indian languages account for a large share of image and photo translation searches, from Hindi signage and Sanskrit texts to Kannada, Tamil, Telugu and Gujarati documents. The editor’s supported font set includes dedicated Noto fonts for Devanagari, Tamil, Bengali, Kannada, Malayalam, Telugu and Gurmukhi, so translated text renders in the right script.
Arabic, Hebrew, Urdu and Persian
Right-to-left scripts need more than swapped words: the translated text has to be set in the correct direction, and mixed lines (an Arabic sentence with an English brand name) have to keep their segments in order. Smartcat re-renders right-to-left text back into the image, with Noto fonts for Arabic, Hebrew and Nastaliq Urdu in the supported set.
Spanish, French, German and other European languages
Tagalog, Amharic, Somali, Burmese and more
Support for less widely spoken languages varies a lot between translation tools, especially for text inside images. If the language you need isn’t listed here, the general picture translator accepts any supported language pair.
Image translation use cases
Who translates images, and why? The answer shapes which tool and which pipeline you should pick.
Marketing and e-commerce visuals
Campaign banners, social graphics, product photos with labels and packaging artwork all carry text that has to land in every market at the same time as the launch. This is where the generative “Marketing & Visuals” pipeline, batch uploads and a shared glossary pay off: one upload, every language, the same product name on every image.
Training and eLearning materials
Annotated screenshots, diagrams and infographics are the parts of a course that usually get left in English. The zero-touch image localization update added SCORM support: upload a SCORM package and images with text are identified, translated and repacked with the course structure intact.
Product documentation and software screenshots
Help-center articles and user guides are full of UI screenshots. Translating them with the same glossary as the interface keeps button labels consistent between the product and the docs. Screenshots are one of the strongest use cases for AI image translation because the source text is clean.
Scanned records, forms and paperwork
Supplier contracts, business licenses, customs paperwork, certificates and archived records often exist only as scans. OCR plus AI gets you a readable, editable draft quickly; for anything legal or official, route the draft to a human reviewer and keep the source text visible so they can check what the OCR read.
Personal and travel translation
Reading a menu, a sign or a label right now is a job for a free camera app — it’s faster and it’s the right tool. Use an image translator like Smartcat when you need the translated image file back: something you can send, file or publish.
| Need | Free camera / lens app | Smartcat image translation |
|---|---|---|
| Read one sign or menu now | Ideal | Works, but more than you need |
| Get a translated image file back | Overlay only | Same format, layout intact |
| Translate a batch of images | One at a time | One project, one download |
| Consistent brand terms | No glossary | Glossary & translation memory |
| Human review | No | In the same workspace |
Images rarely travel alone. If yours live inside PDFs, the PDF translation guide linked below explains how image-heavy PDFs are handled, including translating a PDF with images, where text baked into a diagram is recognized, translated and written back into the picture.
Image translation guides, how-tos & news
Step-by-step guides on translating images and screenshots, and the product updates behind Smartcat’s image translation tools.
Smartcat product documentation
Step-by-step help articles for translating images in Smartcat.