An image is not a PDF with a text layer
There is nothing to extract until a recognizer runs. The deterministic trial does not cover that call.
If you need the words, turn on image parsing for the workspace. The caller still does not name an internal model.
Warnings are part of the result
A blurry photo and a clean screenshot do not deserve the same confidence. The job's warnings are how the next step knows.
Do not fill missing boxes with zeros. Omit geometry the route did not measure.
Index the text you can check
Keep the image id with the text so a reviewer can open the source. A citation to a screenshot without the file is not checkable.
Diagrams that are mostly pictures may return little text. That is a warning, not a reason to invent a description in the parser.
Which image path the workspace can run
A trial workspace rejects an image source. The check includes the route image-vision and file names ending in png, jpg, jpeg, webp, tiff, bmp, gif, or avif. The response is 403, with the message that trial workspaces use deterministic parsing only.
When image parsing is allowed, a standalone PNG, JPEG, or WebP can be recorded as route image-vision. A picture embedded in a PDF is different: the text-layer path stores an image asset and a block whose warning says PDF embedded image extraction stores metadata only in v1. That block does not contain recognized words.
A larger standalone image is sent to the container instead of the image-vision route. On that path the page size is in pixels, EXIF orientation is applied, and the asset id is asset-000001. Recognized text is a paragraph with confidence 0.5 and the warning Tesseract OCR confidence is approximate in v1. If recognition returns nothing, the warning is Image OCR produced no text, and the block is an image reference rather than a paragraph of words.
- Do not send a screenshot on the free trial.
- Read metadata.route before trusting text from an image.
- A PDF image asset without a vision route has no transcribed text.
- Keep the source file id next to any words you index.
The file to send
The trial rejects a PNG, JPEG, WebP, TIFF, BMP, GIF, or AVIF upload with HTTP 403. The message says trial workspaces use deterministic parsing only. Send a text-based PDF of the same page for this first call.
A raster needs an approved paid workspace. When that workspace allows it, read metadata.route before you index any words. A picture embedded in a PDF on the text-layer path stores metadata only and does not contain recognized text.
- The curl below attaches document.pdf, not an image.
- A 403 on an image name is the trial boundary.
- Keep the source file next to any words you later index from a paid image route.
export DOCPARSE_URL="https://docparse.genedai.me"
export DOCPARSE_API_KEY="your_key"
curl -fsS -X POST \
"$DOCPARSE_URL/v1/parse?wait=true" \
-H "Authorization: Bearer $DOCPARSE_API_KEY" \
-F "file=@document.pdf" \
-F "mode=both" \
-o result.json
jq -r '.result.markdown' result.json
jq '.result.document.schemaVersion, .result.document.pages[0].blockIds' result.json
jq '.result.manifest.decision.route, .result.manifest.parserVersion' result.json
Frequently asked questions
Questions this guide answers
Which image types can be parsed?
PNG, JPEG, and WebP can take the image-vision route when the workspace allows image parsing. TIFF, BMP, GIF, and AVIF are also rejected on the trial because the source is an image. Recognition itself requires an approved paid workspace.
Does the free trial read screenshots?
No. A trial workspace returns 403 for an image upload. The error says trial workspaces use deterministic parsing only, and that AI image, Fast, and layout-aware processing require an approved paid workspace.
Should the parser describe a diagram?
An embedded PDF image is stored as metadata in v1, with a warning that says so. A description belongs to a later step, after an image route the workspace is allowed to run.