In short. Grok can inspect images encountered during Web Search when enable_image_understanding is set to true.
Definition: Yes. xAI's Web Search documentation says that setting
enable_image_understandingto true gives the agent access toview_image, which can analyze images encountered during the search process. The documented behavior belongs to xAI's API Web Search tool.
Key facts
- xAI describes Web Search as a tool that searches the web in real time, browses web pages and extracts relevant information.
enable_image_understandingis the switch that gives the agent access toview_image, according to the same Web Search documentation.- When the switch is enabled,
response.server_side_tool_usagereportsSERVER_SIDE_TOOL_VIEW_IMAGEand the number of calls. - The setting also enables image understanding for X Search when X Search is included in the request.
- xAI documents
enable_image_searchas a separate option for finding images and placing them in the response as Markdown image embeds. - xAI's Image Understanding documentation says that some models accept images as input and use their context when generating a response.
How does image inspection work in Web Search?
xAI's Web Search tool lets Grok search the web in real time, browse pages and extract information. With image understanding enabled, the agent has access to view_image and can analyze images it encounters during that search process. The documentation exposes the server-side tool usage field so an API client can see whether the image tool was called and how many times.
The setting appears inside the web_search tool configuration. xAI shows it in examples for the xAI SDK, the Responses API and the Vercel AI SDK. A minimal Python-shaped excerpt matching the xAI SDK example is:
tools=[
web_search(enable_image_understanding=True),
]
The same page says that the Web Search setting also enables image understanding for X Search when that tool is included in the request.
Is image search the same feature?
No. xAI's Web Search documentation describes enable_image_search as a way for Grok to search for relevant images and include returned images in the response as Markdown image embeds such as . It says those returned images are added to the model context used to write the response.
The documentation assigns a different role to enable_image_understanding: it lets Grok inspect images found while browsing regular web pages. A page image encountered during page browsing therefore falls under image understanding, while an image-search result is handled by the separate image-search option.
What does xAI document about direct image inputs?
The separate Image Understanding page describes direct image input rather than an image encountered through Web Search. It says the request is similar to a text-only prompt, with content sent as a list of objects. Its example uses an input_image object with an image_url and an input_text object with a question.
The image_url value can be a public Internet URL instead of a base64 data URL. The page lists a maximum image size of 20MiB, no limit on the number of images, and jpg/jpeg or png as supported image file types. It also says that any image and text input order is accepted, including a text prompt before an image prompt.
Related terms
FAQ
Can Grok inspect every image on a webpage?
xAI documents analysis of images encountered during the search process and exposes the number of view_image calls. The documentation does not specify a rule for which images on a page are selected.
Should I enable image search for a page image?
Use enable_image_understanding for images Grok encounters while browsing regular web pages. enable_image_search covers a separate workflow in which Grok searches for relevant images and includes them as Markdown embeds.
Can I send Grok an image URL directly?
Yes. xAI's Image Understanding documentation says that image_url can contain a public Internet URL or a base64 data URL. This direct input path is separate from the Web Search tool's view_image path.
Sources
Check how a page reads to answer engines. Open the GEO score analyzer