图片翻译器 - OCR 识别并翻译任意语言的照片

两个引擎串联工作：OCR 从照片中提取原始文字，随后大语言模型翻译器结合上下文将其转换为您的目标语言。习语、菜单项、品牌名称和简短标牌都能翻译得干净利落。对于较长的文学段落，可期待与 DeepL 或 Google Translate 等顶级消费级翻译器相当的质量。

可以。英语、西班牙语、法语、德语、意大利语、葡萄牙语、荷兰语、俄语、波兰语、土耳其语、阿拉伯语、希伯来语、印地语、孟加拉语、泰米尔语、中文（简体和繁体）、日语、韩语、越南语、泰语和印度尼西亚语均支持双向翻译。可在法语 → 英语、日语 → 西班牙语、中文 → 印地语以及所支持语言集中的任意其他语言对之间进行翻译。

工整的印刷体手写字和黑板菜单大多能够识别——本 OCR 针对日常标牌和随意手写做了优化。对于连笔字、密集的手写信件或医生处方，请先使用我们的手写 OCR 工具提取文本，再粘贴到翻译器中，以获得更高的准确率。

可以。一次上传即可触发 OCR、语言检测、翻译和并排显示。你还可以在左侧看到原文、右侧看到译文，这对想要对比措辞的语言学习者很有用。点按某个单词即可听到它在目标语言中的发音。

上传的文件会在几分钟内处理并自动删除。翻译记录不会保存在服务器上——你的上一次结果仅保留在浏览器标签页中，关闭页面后即清除。这让该工具可以放心用于旅途突发情况、医疗说明书和私人信函的翻译。

可以。复制按钮让你一键获取原文、译文或两者并排的内容。下载按钮可导出用于记笔记的纯 .txt 文件，或便于与同事分享的简洁 .docx 文件。旅行者把它粘贴到 WhatsApp；学生把它粘贴到 Notion；机构把它粘贴到翻译记忆工具中进行整理。

It swaps the underlying text-recognition model used before translation. Default works well for most printed signs and menus; Engine 1 and Engine 2 are alternates that can read stylized fonts, low-contrast photos or scripts the default struggles with more accurately. If the detected text looks wrong before it's even translated, switching engines is the first thing to try.

Target-language changes re-translate automatically, but source language and engine changes require pressing the Re-run button — an amber notice appears telling you settings have changed. This is deliberate: OCR is a metered step, so the tool won't silently re-scan the image every time you touch a dropdown, only when you confirm a fresh read.

Yes. Once a translation is ready, a toggle button switches the result panel between the detected source text and the translated text, and each version has its own copy button. This helps language learners check how a phrase maps between languages, or verify the OCR read the sign correctly before trusting the translation.

Set the Source language dropdown to the actual language on the sign rather than leaving it on English, since OCR accuracy depends on matching the character set. Then try Engine 1 or Engine 2 if Default misreads stylized lettering — each engine has different strengths on decorative fonts and low-contrast photos. Getting the OCR text right matters more than the translation step, since a bad OCR read guarantees a bad translation regardless of target language.

Image language covers every script our OCR engines can read — around 30 languages depending on the OCR Engine you pick. Translate to draws from a shorter list of roughly 20 output languages that the translation backend supports. That means you can photograph text in almost any supported language, but the translated result can only come out in one of the languages shown under Translate to.

Yes — both the detected-text and translated-text boxes are editable, so you can type a correction directly into whichever one the toggle button is currently showing. Copy always grabs the text you're viewing at that moment, but Download .txt always saves the translated-text version specifically, even mid-edit on the original side — toggle to the translated view first if you want a source-text fix to carry into the downloaded file.

The upload accepts PNG, JPG, WEBP, GIF, TIFF, BMP, HEIC and HEIF, so an iPhone photo in its native HEIC format works without converting it to JPG first. Whichever format you use, set the Image language and OCR Engine to match the text in the photo before running OCR for the most accurate read.