← Back to blog

Photo Mode: Translate Images, Signs and Screenshots

News written by Leo 5 min read

Fink can now translate the text in an image. You upload a photo or screenshot, or take a picture with your phone, and Fink reads the text, translates it and paints the translation over the original. Photo mode covers the same 24 languages as text translation, and we tested it most on Japanese signs and menus.

The back of a Japanese sencha bag before and after photo mode: brewing steps, a vertical line about the harvest, a best-before date and a label table, each covered with its English translation, such as "Put 3 tsp of tea leaves into the teapot." and "Origin: Shizuoka".
The back of a sencha bag from Shizuoka. Fink rebuilds the label as a table and leaves the date as it is. One cell, the storage note, didn't fit its row, and part of the Japanese still shows next to "Net weight".

How to translate a photo or screenshot

On a computer, click the image button at the bottom right of the text box ("Translate an image"), drag a file onto the box, or paste a screenshot with Ctrl+V. On a phone, tap the camera button and pick "Take photo" or "Choose image". The image takes the place of the text box, and Fink sends nothing until you press Translate.

Fink on a phone: the "Translate an image" sheet with the options "Take photo" and "Choose image", and the translated street plate in the result card, with "Text in the image" and "Back Translation" below it.Fink on a phone: the "Translate an image" sheet with the options "Take photo" and "Choose image", and the translated street plate in the result card, with "Text in the image" and "Back Translation" below it.
On a phone, the camera button opens this sheet. The result card shows the translated photo.

Photo mode needs a free account. Without one you can still translate typed text as a guest.

What you get back

Fink covers each piece of text with a box in the color of the surface around it and writes the translation inside. It also reads vertical Japanese, like the column of text on the right of the tea bag above. Press and hold the image to see the original, tap a shortened box to read its full text, and use "Save image" to download the translated picture or share it from your phone.

Below the image, "Text in the image" lists every line with its source and translation, plus a Copy all button. When you select a line, its box lights up on the picture.

Photo mode in the Fink app, Japanese to English: on the left the photo of a street plate reading 自転車等放置規制区域 and 豊橋市, on the right the same photo with "Bicycle etc. No Parking Zone" and "Toyohashi City" painted over the Japanese text.Photo mode in the Fink app, Japanese to English: on the left the photo of a street plate reading 自転車等放置規制区域 and 豊橋市, on the right the same photo with "Bicycle etc. No Parking Zone" and "Toyohashi City" painted over the Japanese text.
A bicycle no-parking plate in Toyohashi, translated in Fink. Photo: Tien Nguyen on Pexels.

The Insights panel treats the result like any other translation: meaning notes on tone and wording, a back translation into the source language, and, when the translation is in Japanese, the pronunciation in Hepburn romanization. The back translation helps most with your own screenshots, such as a chat reply you want to send in a language you can't read.

Try Fink for your next translation

See the translation, every change, and the why on one screen.

Open App

Why Fink translates the whole image at once

Signs and menus hold short phrases with no sentence around them. A Hacker News user noticed that Google Translate's camera mode handles long dialogue well, but "the translation does start to fail with short snappy 'action manga' japanese with short context-less sentences and exclamations."

Mistral OCR reads the text first. Fink then sends all of it to the model in a single request, so the model translates each phrase with the rest of the picture in view and leaves prices and numbers as they are.

Limits of the first version

Photo mode takes one image at a time, up to 10 MB. HEIC photos from an iPhone can fail in desktop Chrome, so convert those to JPEG first. Fink translates up to 2,000 characters per image and paints at most 25 boxes; any further text appears only in the list. PDFs don't work yet. The quality check and Rework are off for images, and image translations don't show up in your history.

Close-ups give the best results. If Fink can't tell where the text sits in a photo, it puts the translation in the list and suggests you take the photo closer to the text.

Reading the image costs a fixed 0.015 credits on top of the translation itself, so a photo uses a little more than the same text typed in. The free plan's 3 credits a month still cover dozens of images, and the pricing page lists the paid plans.

Questions about photo mode

Can I translate a screenshot? Yes. Paste it with Ctrl+V on a computer, or pick it from your gallery with "Choose image" on a phone. Chat screenshots and app screens tend to work well because their text is sharp and straight.

Does photo mode work on my phone? Yes. Open Fink in your phone's browser, tap the camera button and take a photo of the sign or menu.

How is photo mode different from Google Lens? Google Lens is free and translates live through the camera, which suits a quick look at a sign. Fink works on a photo you've taken, translates all its text in one request, and adds meaning notes on tone and wording. Our post on translation in 2026 explains why most people still reach for Google's tools.

Which languages does photo mode support? The same 24 languages as text translation. Fink doesn't detect the language in the image, so set the source language before you press Translate.