Mafdet AI Help Center中文

Web search & image understanding

Two Playground capabilities that are often confused with each other.

Enable it in the Tools menu: sending a message first runs a web search and injects the result summary as context for that turn, with the sources listed underneath.

  • Included with your plan — available from Starter upward; the Free plan does not include it, and each plan has its own monthly search quota. See Subscription plans.
  • basic search depth, up to 3 results
  • Returns title / URL / snippet / source / time only
  • Rate-limited; results are not logged and not stored locally
  • Turning it off stops injection immediately

If web search is not part of your plan, the request returns WEB_SEARCH_NOT_ENABLED.

Image understanding (vision)

This is no longer a beta capability. Image input is a standard feature of the Gemini model families: attach an image and the model sees it directly.

  • Available on any plan that can use the model in question — see Subscription plans
  • Up to 3 files per message; native media (image / audio / video) sent inline is capped at 15 MB per file
  • jpg / png / webp and the other formats listed in File upload
  • Images are not persisted, not logged, and not kept in the session or local storage

If you pick a model that cannot read images, the Playground blocks sending and asks you to remove the attachment or switch model — it never silently falls back to OCR-ing the image into text.

Some models also accept PDF, audio and video natively; the model picker shows what each one supports.

Next steps