Long screenshots to Markdown: what two samples actually showed

The short answer

Keep returned readable descriptions and summaries for both images, rather than a full transcript. ScreenshotsTo returned a service error on both attempts. Long2Text returned copyable transcript previews, but those previews still needed structural cleanup.

This is a small workflow check, not a leaderboard. Two English demo images cannot establish which product is most accurate, fastest or best for every document.

Try the exact same inputs

We submitted the original JPEG files without resizing or cropping them differently for each tool. Neither file contains a private customer conversation.

Keep: a description, not the original wording

We used Keep’s public screenshot-to-Markdown converter, without signing in, once per image. Both conversions finished and produced Markdown.

• Conversation: the output grouped the subject matter into themes and described bubbles and illustrations. It did not reproduce the sequence of messages or identify their speakers.
• Article: the output summarized the sections and described the illustration. It did not preserve the original paragraphs. It also changed the caption’s wording from “ten screens tall” to “ten times tall”.
• Pictures: both outputs described visual elements; neither contained Markdown image references to extracted image files.

These outputs may be useful as an overview. They did not satisfy this test’s requirement to preserve the source text. We did not test other Keep workflows or crop variants, and did not verify the separate downloaded file.

Chat output (.md)Article output (.md)

ScreenshotsTo: no output to score

We uploaded each file to ScreenshotsTo’s Markdown tool and selected Generate Markdown. On each attempt, the page displayed:

Unable to reach the AI service.

We did not receive Markdown. This is a record of two unsuccessful attempts in our browser session, not a diagnosis of the cause or a claim about the tool’s general availability or recognition quality. No accuracy score or export verdict is possible from these attempts.

Long2Text: a transcript preview with cleanup still needed

We uploaded the same files through the live site in fresh anonymous browser sessions, with optional AI proofreading off. We tested the free preview only: no login or purchase, and no access to the paid remainder.

• Conversation: the copied preview retained the opening message sequence and used Maya/Me labels. But it incorrectly treated the date divider as a message from Me. A speaker label in Markdown is not proof that attribution is correct.
• Article: the opening paragraph was transcribed, but the wrapped title was split between a heading and body text, with the byline joined to the second line. The title needs manual repair.
• Pictures: these copied previews contained image placeholders, not usable Markdown image links. That does not establish the behavior of a paid image bundle.

The previews contained 783 characters for the chat and 396 for the article, before adding a final newline to the evidence files. Their cutoffs are an access limit, not evidence that the OCR omitted the remainder. Full-text accuracy, the ending of each fresh conversion and paid exports remain untested here.

Copied chat preview (.md)Copied article preview (.md)

Our guides also offer a recorded chat example and a recorded article example, including downloadable bundles. Those are cached examples, not fresh full outputs from this comparison. Check current access and pricing before converting your own files.

What we measured—and what we did not

• One conversion per image per tool on September 12, 2026. No repeated runs to select a better answer, private images, paid upgrades or quota workarounds.
• Keep files preserve the displayed Markdown text; Long2Text files preserve the copied free preview. Only a terminal newline was added where needed. ScreenshotsTo produced no output artifact.
• We visually checked source wording and structure against the available output. There is no manually verified full reference transcript, so we do not report character-accuracy percentages.
• No speed ranking: we did not capture comparable conversion timing boundaries across all three tools. Upload, processing, preview and paid export are different milestones.
• Optional Long2Text proofreading was off. We did not establish equivalent models, settings or preprocessing in the other tools.
• We did not verify an Obsidian import, paid downloads, handwriting, formulas, complex tables or multilingual accuracy.

How to judge your own screenshot-to-Markdown result

1. Check wording before appearance. Compare a sentence near the top, middle and end. If you only have a preview, you cannot check the full document yet.
2. Check structure. Look for date dividers assigned to speakers, split titles, repeated lines and merged paragraphs. Review names and numbers especially carefully.
3. Check the actual export. A picture description or an image placeholder is not an image file. Open the Markdown in your destination editor and verify linked images before relying on it.
4. Match the tool to the job. A summary can help you understand an image; a transcript is needed to reuse its wording. When original text is available, copy or export it instead of introducing OCR errors.

For the next step, use the screenshot-to-Markdown workflow or the chat transcript guide. Keep the source image alongside any text you need to verify.