How to Translate Text in an Image or Photo

OCR guide · 5 min read

You're standing in front of a menu in Lisbon, or holding a product label in Korean, or looking at a street sign that might be pointing you toward the train station. You pull out your phone or point at your screen, try to select the text so you can look it up, and nothing happens. That's because it isn't text at all. It's a picture of text, locked inside an image file.

The good news: getting a translation is a simple two-step process once you know why the obvious approach fails. This guide walks through it, and shows where a free tool like Textquill fits in.

Why you can't just copy text from a photo

A photo of a menu is a grid of pixels, not characters. Your device sees colored dots, not the words bacalhau à brás. There's nothing to highlight and nothing to copy, so you can't paste it into a translator. That's the first wall.

The second wall is script. If the image is in Hindi (Devanagari), Chinese (Han), Arabic, Thai, or Nepali, even reading it aloud to type into a translator is out of reach for most people. You can't retype characters you don't recognize.

To translate an image, you first have to turn the picture of text back into real, editable text. That's what optical character recognition (OCR) does, and it's the missing step most people skip.

The two-step fix: extract, then translate

Here's the honest workflow. It has two distinct stages, and no single button does both today:

1. Extract the text from the image with OCR, so you have real, copyable characters in the correct script.
2. Translate that extracted text using a translation service of your choice.

Textquill handles the first stage. It's a free Chrome extension that runs OCR on your device and pulls the text out of an image in any of 16 languages, including English, Hindi, Chinese, Japanese, Korean, Arabic, French, German, Spanish, Italian, Portuguese, Russian, Thai, Vietnamese, Turkish, and Nepali, with an auto-detect option. To be clear: Textquill extracts the text, it does not translate it. The translation happens in step two, in a separate tool. (On-device AI translation is on Textquill's roadmap, but today it's extract-then-translate.)

Step 1: Extract the text with Textquill

Set the recognition language to match the image

Before you extract, tell Textquill what language the image is in, or leave it on auto-detect. This matters more than anything else. OCR reads shapes, and if you ask it to read a Devanagari or Han image as English, you'll get garbage, random Latin letters that mean nothing. Matching the source script is what makes the extraction accurate.

Grab the text

There are three ways to feed an image to Textquill:

• Right-click an image on a web page and choose the Textquill option, useful for online menus, forums, and product pages.
• Press Alt+Shift+S to draw a box over any area of your screen, handy for a foreign sign in a video or a caption in a screenshot.
• Upload a photo you took, such as a picture of a menu, a label, or a document.

Textquill recognizes the characters and hands you clean, editable text you can copy.

Step 2: Translate the extracted text

Now that the words are real text, translating is trivial. Copy the extracted text and paste it into whatever translator you prefer:

• Google Translate
• DeepL
• Any translation site, app, or built-in browser feature you already use

Pick the target language, and you have your answer. Because the source is now plain text, you can also fix any small OCR slip, a stray character, a missed accent, before you translate, which keeps the result clean.

Why extracting first beats a blurry auto-translate overlay

Camera apps that paint a translation directly over a live image are convenient, but they translate their best guess of the text, and that guess is often wrong on a curved menu, a stylized sign, or in dim light. You never see the source, so you can't tell whether the translation is off because the wording is odd or because the OCR misread it.

Extracting first gives you control at every step:

• Clean source text produces a better translation. Translators do their best work on accurate input.
• You can proofread before translating. Spot a wrong character? Fix it, then translate.
• You keep the text. Save it, search it, or re-translate it into another language later.

Real situations where this helps

• A menu abroad: extract the dishes, translate, and know what you're ordering.
• A foreign document: pull the text from a scanned contract or form, then translate the parts you need.
• A product label: extract ingredients or instructions in another language and check them.
• A street sign or notice: screenshot it, extract, translate, and find your way.
• A message in another language: lift the text from an image someone sent you and read it.

Tips for accurate extraction

• Match the source language or script. This is the single biggest factor. Auto-detect handles most cases, but set it manually if you know the language.
• Crop tight. Select only the text you care about, not the whole photo. Less clutter means fewer mistakes.
• Favor good contrast. Dark text on a light background (or vice versa) reads far better than low-contrast or glare-heavy shots.

One more thing worth knowing: Textquill's OCR runs on your device, so the image you're reading, whether it's a private document or a personal message, doesn't get sent off to a server just to be extracted.

FAQ