One Reading Goal, Four Media

FluentRead’s context-preserving approach becomes harder—and more revealing—when the source is not ordinary webpage text. This chapter compares four surfaces: images, selected screen areas, documents, and video. The useful choice is not based on a long format list; it depends on what can be captured, where recognition happens, what text crosses into translation, and how the result can be checked against the source.

Images and areas: preserve the source while recognizing its words

Image translation is for text embedded in a webpage image: a chart label, a comic panel, a scanned illustration, or a screenshot. The extension recognizes the text locally, sends the recognized text to the selected translation service, and can overlay the result on the image. A separate text view exposes the recognized and translated content for copying. The original image remains available, so the reader can compare names, numbers, units, and layout instead of treating the overlay as a new authoritative image.

That sequence has an important privacy and quality boundary. The pixels used for recognition do not have to be uploaded for text translation, but the recognized text can still leave the browser. Local recognition therefore reduces one kind of exposure without eliminating the provider request that follows. It also creates a failure mode before translation begins. If a decorative font, vertical layout, complex background, table, or blurry source causes recognition to miss a character, a later translation model receives the wrong input. A fluent result cannot prove that the original character was read correctly.

The first use of a new source language may require a recognition pack. Automatic detection prepares a limited set of common languages, while other languages require the reader to choose the source language and download its pack. This is not the same dependency as a translation service. A translation provider can support a language pair while the local recognition stage lacks the language data needed to produce text in the first place. Choosing the source language and preparing the recognition pack are therefore part of making the image capturable, not merely part of configuring the final translation.

Image translation is most useful when the image itself is the context the reader needs to preserve. An overlay can keep the translated label near the original position, which matters for diagrams or panels whose meaning depends on spatial arrangement. The text view is better when the overlay becomes too small or when the reader needs to copy a long result. Switching between original and translated image provides a reversible check, but the reader should still inspect difficult names, figures, and units.

Area translation starts from a different problem. The reader may not want to process an entire image, or the text may be inside a screenshot, a paused video frame, a chart, or a comic bubble that the image hover entry does not expose. The reader invokes the area shortcut and draws a crop.