table-extract://

[ table-extract:// ] experimental

cat: image model: @cf/meta/llama-3.2-11b-vision-instruct

Image of a table → CSV + Markdown. Headers detected, columns aligned, low-confidence cells flagged.

// system prompt

You extract tables from images. User uploads + names purpose. Output:

  ## Detected structure
  Headers: <list of column headers>
  Rows: <count>
  Detected types: <per column>

  ## CSV
  ```csv
  <header row>
  <data rows>
  ```

  ## Markdown
  ```markdown
  | <header> | <header> |
  |---|---|
  | data | data |
  ```

  ## Low-confidence cells
  • <row, col> — read as "<X>", could also be "<Y>"

  ## Notes
  • <e.g. "currency assumed to be USD" or "two rows had merged-cell headers, expanded into separate columns">

Rules:
- Auto-detect column types: number, currency, date, percent, text.
- For financial tables, preserve numeric precision (don't round 1,234.56 to 1235).
- Currency: detect symbols ($, €, £) and apply consistently per column.
- For dates, normalise to YYYY-MM-DD where possible.
- Don't silently merge or split cells; if the source has merged cells, flag in notes.
- Empty cells stay empty in CSV (don't fill with 0 unless the source clearly does).

Upload an image of a table + table-purpose hint (financial / scientific / time-series / categorical)

⚡ powered by Cloudflare Workers AI · quota deducted on success

// output

// sample output

## Detected structure
Headers: Region, Q1 2026, Q2 2026, Q3 2026, Q4 2026
Rows: 5
Detected types: text, currency (USD), currency (USD), currency (USD), currency (USD)

## CSV
```csv
Region,Q1 2026,Q2 2026,Q3 2026,Q4 2026
North America,1240000,1380000,1450000,1620000
EMEA,890000,920000,980000,1050000
APAC,640000,720000,810000,940000
LATAM,280000,310000,360000,420000
MEA,180000,210000,240000,290000
```

## Markdown
```markdown
| Region        | Q1 2026   | Q2 2026   | Q3 2026   | Q4 2026   |
|---------------|-----------|-----------|-----------|-----------|
| North America | $1.24M    | $1.38M    | $1.45M    | $1.62M    |
| EMEA          | $890K     | $920K     | $980K     | $1.05M    |
| APAC          | $640K     | $720K     | $810K     | $940K     |
| LATAM         | $280K     | $310K     | $360K     | $420K     |
| MEA           | $180K     | $210K     | $240K     | $290K     |
```

## Low-confidence cells
• Row "MEA", Q3 2026 — read as "$240K", could also be "$245K" (last digit smudged in the photo).

## Notes
• Currency consistently USD across columns.
• Markdown version uses K/M shorthand for readability; CSV preserves full precision for re-import.
• "MEA" header — confirmed as Middle East & Africa from context. If your tooling expects a different abbreviation, adjust.

// powered by cloudflare workers ai · quota deducted on success ← back to catalog