---
title: "When should I enable vision correction?"
description: "How vision correction rebuilds layout with a multimodal model, how on and off compare, and which document types suit each setting"
canonical_url: "https://docs.ardge.com/en/knowledge/precisemd/vision-correction"
---
# When should I enable vision correction?

> How vision correction rebuilds layout with a multimodal model, how on and off compare, and which document types suit each setting

## Short answer

Turn **Vision correction** on for documents with complex layout: tables, multi-column pages, reports, and forms of every kind all need a multimodal model to rebuild their structure from the original image. Turn it off for plain-text documents with simple layout, and save roughly 5 seconds per page. The setting defaults to on in both **Playground** and **OCR tasks**.

## How it works

With **Vision correction** off, the document only passes through the OCR model, and the recognized text is written straight into Markdown. That's the fastest path, but the layout information is limited to what the OCR model itself made of the page.

With **Vision correction** on, after OCR the system has a multimodal model read the page's original image, correct misrecognized characters and digits against it, and rebuild tables as HTML tables that preserve merged cells. The document description you write in the prompt field is passed to that model as context.

## Comparison

<table>
<thead>
  <tr>
    <th>
      
    </th>
    
    <th>
      Vision correction off
    </th>
    
    <th>
      Vision correction on
    </th>
  </tr>
</thead>

<tbody>
  <tr>
    <td>
      Processing time per page
    </td>
    
    <td>
      Baseline
    </td>
    
    <td>
      About 5 seconds more
    </td>
  </tr>
  
  <tr>
    <td>
      Table structure
    </td>
    
    <td>
      Whatever the OCR model made of it
    </td>
    
    <td>
      Rebuilt from the original image, merged cells included
    </td>
  </tr>
  
  <tr>
    <td>
      Prompt
    </td>
    
    <td>
      The field is disabled — no input
    </td>
    
    <td>
      Available; the closer the description, the more accurate
    </td>
  </tr>
  
  <tr>
    <td>
      Models
    </td>
    
    <td>
      OCR model only
    </td>
    
    <td>
      Plus a multimodal model
    </td>
  </tr>
  
  <tr>
    <td>
      <code>
        enable_vlm
      </code>
      
       in the output
    </td>
    
    <td>
      <code>
        "false"
      </code>
    </td>
    
    <td>
      <code>
        "true"
      </code>
    </td>
  </tr>
</tbody>
</table>

## Which one should I use?

- **Turn vision correction on** if the document has tables, multi-column layout, reports, or forms, or if you need the output to preserve the original layout structure
- **Turn vision correction off** if the document is single-column plain text and total batch processing time matters to you
- **Run both settings once in Playground** if you can't tell. Use the **Diff** tab under **Compare** to see the same file's output under each setting, then decide what your **OCR task** should use

## Limits and caveats

- **The two settings are independent.** **Playground**'s **Advanced settings** and the **OCR task** form's **Advanced settings** each have their own **Vision correction** toggle; what you tune in Playground doesn't carry over to the task, so set it again on the task form
- **You can't write a prompt in advance without it.** The prompt editor is disabled while **Vision correction** is off, showing "Enable vision correction first to enter a prompt."
- **Office formats are unaffected.** DOCX, XLSX, and PPTX go through the conversion pipeline, skipping OCR and vision correction entirely, and their output's `enable_vlm` is always `"false"`
- **Turning it on adds a dependency.** The correction stage sends a request to the multimodal model; when that model can't respond, the file is marked failed with "The AI service is temporarily unavailable. Try again later"

## Related articles

- [Tune prompts and compare OCR results in Playground](https://docs.ardge.com/en/knowledge/precisemd/playground-prompt-tuning)
- [Schedule an OCR task to process a folder automatically](https://docs.ardge.com/en/knowledge/precisemd/schedule-ocr-tasks)
- [What files are produced in the OCR output folder?](https://docs.ardge.com/en/knowledge/precisemd/ocr-output-files)


## Sitemap

See the full [sitemap](https://docs.ardge.com/sitemap.md) for all pages.