Result Formats & Output

Result Formats

The result format determines how the AI returns extracted data. Choosing a format is the first step when creating a template — it controls which configuration options are available and how the AI processes your document.

Choosing a Format

JSON

The default format. The extracted data is returned as a JSON object matching your schema. Use this when integrating with APIs, storing in databases, or building custom processing pipelines.

Example output:

{
  "invoice_number": "INV-2024-0042",
  "invoice_date": "2024-11-15",
  "total_amount": 10700,
  "currency": "THB",
  "vendor": {
    "name": "บริษัท เทค โซลูชัน จำกัด",
    "tax_id": "0105567012345"
  }
}

CSV

Tabular format suitable for Excel or Google Sheets. See the CSV Configuration section below for detailed CSV configuration options including column ordering and two CSV modes.

XML

Structured data in XML tags, compatible with enterprise and legacy systems.

Example output:

<?xml version="1.0" encoding="UTF-8"?>
<extraction>
  <invoice_number>INV-2024-0042</invoice_number>
  <invoice_date>2024-11-15</invoice_date>
  <vendor>
    <name>บริษัท เทค โซลูชัน จำกัด</name>
    <tax_id>0105567012345</tax_id>
  </vendor>
  <total_amount currency="THB">10700</total_amount>
  <line_items>
    <item>
      <description>Cloud Server Monthly</description>
      <quantity>1</quantity>
      <unit_price>10700</unit_price>
    </item>
  </line_items>
</extraction>

DOCX and PDF

The AI-extracted data is formatted into a Word or PDF document. Useful for generating human-readable reports that can be printed, shared, or archived without any additional processing.

Example structure of a generated PDF/DOCX:

• Title: Document type and reference number
• Header section: Key metadata (date, vendor, customer)
• Data table: Extracted line items in a formatted table
• Summary: Totals, tax calculations, and notes

Text

Raw text output from the AI. Suitable when you need a plain description, summary, or transcription of the document content. Use with the Free Text extraction mode.

Example output:

Invoice INV-2024-0042 dated November 15, 2024.
 
Issued by บริษัท เทค โซลูชัน จำกัด (Tax ID: 0105567012345) to
the purchasing department.
 
Items:
- Cloud Server Monthly: 1 unit × ฿10,700.00 = ฿10,700.00
 
Subtotal: ฿10,700.00
VAT (7%): ฿749.00
Total: ฿11,449.00
Payment terms: Net 30 days.

HTML

The AI returns an HTML-formatted response. Useful when you want to display results directly in a web interface, embed in emails, or render rich text output.

Example output:

<div class="invoice-extraction">
  <h2>Invoice INV-2024-0042</h2>
  <table>
    <tr><th>Date</th><td>2024-11-15</td></tr>
    <tr><th>Vendor</th><td>บริษัท เทค โซลูชัน จำกัด</td></tr>
    <tr><th>Tax ID</th><td>0105567012345</td></tr>
  </table>
  <h3>Line Items</h3>
  <table>
    <tr><th>Description</th><th>Qty</th><th>Price</th><th>Amount</th></tr>
    <tr><td>Cloud Server Monthly</td><td>1</td><td>฿10,700</td><td>฿10,700</td></tr>
  </table>
  <p><strong>Total: ฿11,449.00</strong> (incl. 7% VAT)</p>
</div>

Image

The AI generates an image based on your document content or instructions. See the Image Generation section below for detailed image generation configuration.

CSV Configuration

CSV output has two modes that give you control over the structure of the exported spreadsheet.

Simple Mode (Columns Only)

Enable csvColumnsOnly to generate a CSV where each row contains only the values listed in csvColumnOrder, without a full schema expansion. This is ideal for flat data where you want a simple column-by-column layout.

Example configuration:

{
  "resultFormat": "csv",
  "csvColumnsOnly": true,
  "csvColumnOrder": [
    "invoice_number",
    "invoice_date",
    "vendor_name",
    "total_amount",
    "currency"
  ]
}

Output CSV:

invoice_number,invoice_date,vendor_name,total_amount,currency
INV-2024-0042,2024-11-15,บริษัท เทค โซลูชัน จำกัด,10700,THB
INV-2024-0043,2024-11-18,ร้านค้าตัวอย่าง,5350,THB

Full Schema Mode

When csvColumnsOnly is false (the default), the full JSON Schema is used to generate the CSV structure. Nested objects and arrays are flattened using dot notation.

Example — a schema with nested vendor object produces these columns:

invoice_number,invoice_date,vendor.name,vendor.tax_id,total_amount

Column Ordering

Use csvColumnOrder to specify the exact order of columns in the output:

{
  "csvColumnOrder": ["invoice_number", "invoice_date", "total_amount", "vendor.name"]
}

Columns not listed in csvColumnOrder will appear after the ordered columns. This lets you pin the most important columns to the left.

CSV Example: Receipt Scanner

Schema:

{
  "type": "object",
  "properties": {
    "receipt_date": { "type": "string" },
    "store_name": { "type": "string" },
    "item_name": { "type": "string" },
    "quantity": { "type": "number" },
    "unit_price": { "type": "number" },
    "total": { "type": "number" },
    "payment_method": { "type": "string" }
  }
}

CSV config:

{
  "resultFormat": "csv",
  "csvColumnsOnly": true,
  "csvColumnOrder": ["receipt_date", "store_name", "item_name", "quantity", "unit_price", "total", "payment_method"]
}

Image Generation

When the result format is set to image, Ocriva uses an AI image generation pipeline instead of text extraction. The AI reads your document content or follows your instructions and generates a corresponding image.

This is useful for:

• Converting text descriptions into visual assets
• Generating product images from specification sheets
• Creating illustrations from written content
• Producing visual summaries of reports

Image Options (imageUserOptions)

Configure image generation behavior with these options:

Aspect Ratio

Controls the dimensions of the generated image:

Art Style

Sets the visual style of the generated image:

Color and Lighting

Negative Prompt

Use negativePrompt to list what the AI should avoid in the generated image. This helps prevent unwanted elements:

{
  "imageUserOptions": {
    "aspectRatio": "16:9",
    "artStyle": "illustration",
    "colorLighting": "bright_airy",
    "negativePrompt": "text, watermarks, blurry, low quality, distorted faces"
  }
}