Screenshot API

1. Save the options

Save from the dashboard under Screenshots, or over the API. Names are unique per workspace, so saving under an existing name edits it rather than creating a duplicate. Options are validated as they are stored — a preset that could never capture is rejected here, not later on reuse.

curl -X POST https://pixbix.app/api/v1/screenshot/presets \
  -H "x-api-key: pk_live_…" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "Pricing page — retina",
    "options": {
      "url": "https://acme.in/pricing",
      "format": "png",
      "fullPage": true,
      "scale": 2,
      "viewport": { "width": 1440, "height": 900 },
      "blockCookieBanners": true,
      "hideSelectors": ["#intercom-container"],
      "waitUntil": "networkidle2"
    }
  }'

2. Capture with it

Send presetId and nothing else. Every option the preset holds is applied, including the URL — so this is a complete capture request.

curl -X POST https://pixbix.app/api/v1/screenshot \
  -H "x-api-key: pk_live_…" \
  -H "Content-Type: application/json" \
  -d '{ "presetId": "prs_9d2f" }'

3. Override what changes

Anything you send alongside presetId wins. The usual case is reusing one recipe across many pages by overriding only the URL.

{
  "presetId": "prs_9d2f",
  "url": "https://acme.in/enterprise",
  "viewport": { "width": 390 }
}

Across a batch

A top-level presetId applies to every item, which is the shortest way to capture a list of URLs under one recipe. An item may name its own preset, which wins for that item.

{
  "presetId": "prs_9d2f",
  "items": [
    { "url": "https://acme.in/pricing" },
    { "url": "https://acme.in/enterprise" },
    { "url": "https://acme.in/startups", "fullPage": false }
  ]
}

In workflows

A capture step’s extracted values land on the row itself, so a later step reads them as {{item.price}} — the same expression it would use for a spreadsheet column. That is what makes a capture a data source: shoot a product page, then render a card from what was on it. A column the row already has always wins, so a mapping you made by hand is never silently replaced.

In image templates

A live-capture layer can carry extractions too, and each one may name a layerId. The captured page then fills that layer on every render: shoot the product page, and the headline beside the screenshot is that product’s name. A value passed in by the caller wins over one read from the page, since an explicit field is a deliberate instruction.

What the layer does with the value depends on what it is. A text layer takes it as words. An image layer takes it as a picture — read the src of the hero photo or the content of og:image, and that layer becomes the product’s own photo at full size, composited into the design rather than cropped out of a screenshot of the page it sat on. A relative address resolves against the page it was read from; anything that is not an http(s) address after that is left alone, and the layer keeps its existing picture.

A data element — a chart, a table, a widget, a QR code — is fed the same way, with layerField naming which part of it to fill. A chart takes values, labels or title, and series:2 when it plots several series and you mean one of them; a table column:0, column:1 and so on; a widget value, label, suffix or max; a code layer value. Omit it and each takes its primary slot.

Because the slot is named, several extractions can feed one element: read the product names with allinto a table’s first column and the prices into its second, from one visit. Quantities are parsed out of whatever the page prints — ₹1,299, 73% and 1.234,50 all arrive as numbers — and anything with no number in it is left out rather than plotted as zero.

A list-shaped slot can also be addressed one entry at a time by appending its position: values:2 is the third bar, labels:0 the first label, and column:1:4 one cell — column 1, row 4. Without an index the whole list is replaced, which is what an all extraction is for; with one, a single figure updates and the rest of the chart stays as it was drawn. An index past the end grows the series to reach it rather than writing nowhere — a bar beyond the last label is drawn under its position number, and a column beyond the last heading is added to the table with a blank one, so a figure that was read is never a figure that goes missing.

Every extraction also comes back on the render itself, under extracted — including the ones that name no layerId and so appear nowhere in the picture. It is on the JSON response, on the render when you fetch it back, and in the render.completed webhook; a binary render carries it in the X-Pixbix-Extracted header instead, since bytes have nowhere else to put it.