Transcribe
🤖 AI Drafted (Not reviewed)
Extract text from images (OCR)
|
|
| Tool id |
transcribe |
| Category |
vision |
| Uses a language model |
yes |
| Needs a generative model |
no |
| Runs over many items |
yes |
| Item handling |
elementwise |
| Structured output |
yes |
| Human-verified |
yes |
What it reads
| Port |
Type |
Required |
What it is |
Files (files) |
files |
yes |
Image files |
Context (context) |
any |
no |
Previous text/transcription |
Metadata (metadata) |
json |
no |
Existing metadata |
Documents (documents) |
json |
no |
Document metadata |
What it emits
| Port |
Type |
Required |
What it is |
Text (text) |
text |
— |
Raw text response |
Value (value) |
any |
— |
Parsed value |
Texts (texts) |
array |
— |
Per-item texts |
Values (values) |
array |
— |
Per-item values |
Results (results) |
json |
— |
Full results |
Records (records) |
array |
— |
Per-document text records [{doc_id, text}, …]. |
Artifacts (artifacts) |
json |
— |
Artifact IDs |
Options
| Option |
Type |
Default |
What it does |
choices |
array |
— |
Valid choices. (Not shown in the editor.) |
chunk_size_chars |
integer |
0 |
Chunk large input text above this character budget (0=auto). |
force_ocr |
boolean |
no |
Force image processing instead of existing text. |
language |
string |
auto |
Language locale, or ‘auto’ to follow the library’s language policy. |
match_mode |
string |
prefer |
Match mode. One of: prefer, strict, inform. |
max_image_dimension |
integer |
8192 |
Max image size. |
max_items |
integer |
10 |
List max items. |
max_tokens |
integer |
8192 |
Max response. |
max_words |
integer |
50 |
Word limit. |
metadata_field |
string |
— |
Save to field. |
model_name |
string |
— |
Model name. |
output_format |
string |
text |
Response format. One of: text, boolean, choice, number, words, list, json. |
prompt |
string |
— |
Custom prompt. |
provider_name |
string |
— |
LLM provider. One of: openai, anthropic, google, ollama, lmstudio, groq, together, deepseek, mistral, openrouter, dashscope, xai, perplexity, fireworks, deepl. |
quality_gate |
boolean |
yes |
Stop the run if output is unreadable. |
reference_values |
object |
— |
Known values to match. (Not shown in the editor.) |
regions_first |
boolean |
yes |
Detect text regions on-device (Apple Vision) BEFORE transcribing, saving per-page bounding boxes as a regions artifact — works for every provider, unlike return_boxes. |
return_boxes |
boolean |
no |
Ask the model for word/line bounding boxes (Gemini only; other providers transcribe without geometry and say so). |
save_to_db |
boolean |
yes |
Save to library. |
save_to_file |
boolean |
no |
Export to file. |
temperature |
number |
0.7 |
Creativity. |
thinking_mode |
string |
off |
Chain-of-thought reasoning depth. One of: off, short, medium, long. |
update_page_content |
boolean |
yes |
Index for search. |
vision_mode |
string |
auto |
Vision engine. ‘auto’ picks based on the resolved provider: apple → Apple Vision OCR; anything else → LLM vision path. One of: auto, apple, llm. (Not shown in the editor.) |
The prompt it sends
This is what the tool asks a model, with every option left at its default. Changing the options above changes this text.
Transcribe the text visible on this image.
Language: transcribe in the language of the source. Do not translate, and do not assume the document is in English.
Rules:
- Output ONLY the transcription. No headings, no preamble, no commentary,
no summary, no notes, no explanations, no observations about quality or
legibility, no descriptions of seals or images, no language about the
difficulty of the handwriting.
- Preserve original layout, line breaks, and paragraph structure.
- Preserve original spelling and capitalisation, including ALL CAPS
headers if they appear that way.
- Preserve orthography exactly as written, including all diacritics
and accent marks (e.g., keep "Chocó" as "Chocó", never "Choco";
keep "Ramón" as "Ramón", never "Ramon").
- Do not strip accents, tildes, cedillas, or umlauts. If a mark is
visible, keep it.
- Include every visible text element — headers, body, marginalia, stamps,
signatures (transcribe the signed name as written), printed labels,
handwritten annotations.
- For text you cannot confidently read, use explicit uncertainty markers:
[ilegible] for unreadable text and [uncertain] for plausible-but-low-
confidence readings. Place the marker inline at the uncertain span.
Do not guess. Do not fill in.
- Do NOT invent dates, numbers, names, or words that are not legibly
present. Do not normalise dates ("23/7/1999" stays "23/7/1999", not
"1999-07-23").
- Do NOT repeat any portion of the transcription. Output each visible
passage exactly once.
- If the image contains no legible text, output the single token
[sin texto].