## Why are long screenshots so hard

## to turn into text?

Long2Text Journal . 5 min read

Phone screenshots keep getting longer: a chat history or a full article can be tens of thousands of pixels tall. Yet the moment you try to pull the text out, most tools fail or silently drop whole sections.

[Illustration: A pixelated landscape with mountains and a moon.]

Figure 1: a long screenshot is often ten screens tall

## Problem one: the image is too tall

Most recognition services first shrink an image to a fixed size. A screenshot 10,000 pixels tall ends up with only a few pixels per letter, and the text becomes unreadable.

Worse, many services cap the upload size, so very long screenshots can't even be sent.

## Problem two: cutting breaks lines

Cutting the image into pieces is the obvious fix, but when a cut lands in the middle of a line, both pieces get half of it. The answer is to overlap neighbouring pieces and remove duplicates by position.

## Problem three: the structure is gone

Even with every letter right, you get a pile of broken lines. Who said what, when, and which line was a heading - all lost.

## What we do

Split the image into overlapping slices in the browser Read all slices at once, and zoom in on tiny text Rebuild paragraphs by position, with speakers for chats Cut out stickers and photos and put them back in place

With that, even a screenshot tens of thousands of pixels tall comes back as a clean, well-formatted document in about fifteen seconds.

## What's still hard

OCR engines are sometimes confidently wrong, especially on a screenshot inside a screenshot. We let Al proofread those spots using the surrounding context, and mark every change so you decide whether to keep it.

Charts and handwriting are the other challenge. They have no clean text, so the best we can do is describe them in a sentence, and that part is still improving.

## Finally

Long screenshots were meant to make sharing easy, yet they often become a dead end for information. We'd like them to be searchable, copyable, editable text again.