Question 1

What does OCR Recognize Text do?

Accepted Answer

It scans an input image and returns every detected text fragment along with its bounding box (as fractions of image size) and a confidence score, without regenerating or modifying the image.

Question 2

What parameters are configurable?

Accepted Answer

The only required input is `image_url`, the public URL of the image to scan. An optional `webhook_url` can be supplied to receive the result asynchronously.

Question 3

Why is `bbox` returned as fractions instead of pixels?

Accepted Answer

Fractional coordinates (0–1 range) let the same bounding box be applied correctly regardless of any resizing that happens between detection and a later edit step, without extra unit-conversion math.

Question 4

What does the `confidence` field mean?

Accepted Answer

Each fragment is labeled `high`, `medium`, or `low` based on the OCR engine's internal recognition score, so downstream logic can choose to ignore low-confidence noise.

Question 5

Can this be used to edit text in an image, not just detect it?

Accepted Answer

OCR Recognize Text only detects and returns text fragments. To change the text, pass the detected fragments and your desired replacements to an image-edit model as a follow-up call.

Question 6

What image formats are supported?

Accepted Answer

Standard web image formats (JPG, PNG, WebP) served from a publicly reachable URL are supported.