Screenshot any web page
web_screenshot is a $0.20 GET request on the Wuthering AI API. Render a page in a real browser and get back an image of it. It answers complete JSON, and a call that fails is not charged.
Endpoint
GET /v1/web_screenshot
Price per successful call
$0.20
Failed calls
Not charged
curl 'https://wutheringai.com/v1/web_screenshot?url=…&token=YOUR_TOKEN'
What does web_screenshot return, and when should you call it?
Render a URL in a real browser — JavaScript executed, resources loaded — and return an image of the result.
This answers what wuthering_web_fetch cannot: how a page LOOKS. Layout, the offer above the fold, which plan a pricing table visually pushes, what a competitor's ad or hero actually shows. Text extraction throws all of that away, and any claim about design, prominence or placement made from markdown is a guess.
Rendering takes up to thirty seconds because the page is genuinely loaded, not fetched. That is the normal duration, not a stall — do not retry a call that is still running.
Sites that fight automation — social networks most of all — refuse a share of renders, and that refusal arrives as an error saying the target URL is invalid even when the URL is exactly right. Read it as 'this page would not load this time', not as a URL you got wrong: retry once before concluding anything, and never rebuild the URL in response to it. Failed calls are not charged, so a retry costs only time.
The renderer's own geography cannot be set, so a page that varies by country may come back localised to somewhere other than the market you meant. Judge layout, prominence and creative from it freely; treat language, currency and regional pricing as unverified.
Returns: an items array holding one image field, whose value is a URL you can open or pass to a vision model. It is served from this API and is short-lived — read it soon after the call, and call again rather than storing it for later.
Use when the visual matters, or when a page's meaning is carried by its layout. Don't use it to read a page's text, which is wuthering_web_fetch at a fraction of the wait, and don't use it to find a page, which is wuthering_web_search.
What parameters does web_screenshot take?
web_screenshot accepts 4 parameters, of which 1 is required. Array values are comma-separated on the query string.
| Parameter | Type | Required | Description |
|---|---|---|---|
| url | string | yes | The full URL to render, including protocol. If you do not have the exact URL, find it with your OWN web search first (for example: site:acme.com pricing), or with wuthering_web_search if you have none. A guessed URL renders someone's 404 page as though it were the page you asked for, and an image makes that harder to notice than text does. |
| device | string | no | Which viewport to render at. Defaults to desktop (1920x1080). mobile renders at 390x844 and is the right choice when the question is about the mobile experience, which is often a different layout rather than the same one narrowed.One of: desktop, mobile, tabletDefaults to desktop |
| full_page | boolean | no | Capture the entire scrollable page. Defaults to true. Set false for just what is visible before scrolling, which is what you want when the question is about the first impression rather than the whole document.Defaults to true |
| hide_cookie_banner | boolean | no | Dismiss the cookie-consent overlay before capturing. Defaults to true, because on most European and many US sites the banner covers the part of the page worth looking at.Defaults to true |
Reference for web_screenshot · last updated · published by Wuthering AI