Find public web images from a text query and return image URLs with their source pages. Use the results to select visuals for documents, feeds, or creative references, with checks for image links used in deliverables.
中文简介正在准备中。当前显示英语简介。
Search the web by text query for image URLs and source pages for feeds, artifacts, and visual references. Does not identify a supplied image or person.
安装前可以查看交给 AI 代理的指令内容。
Use the bundled CLI to find public images by text query:
/opt/hatch/bin/image-search "Golden Gate Bridge at sunset" --max-results 5
Optional result language:
/opt/hatch/bin/image-search "Paris architecture" \
--max-results 5 \
--language fr
The CLI returns structured results and omits unavailable fields. Interpret the URLs as follows:
media_url is the public source image locator and the preferred full-image
URL when present.thumbnail_cdn_url is Meta's renderable CDN preview and the fallback when
media_url is absent. Treat it as a cache URL, not the sole durable copy of
an artifact.media_handle and candidate_ref are internal, non-renderable fields. Do
not fetch, display, or pass them as image URLs.page_url is the source page. Retain it for provenance or attribution.The service returns only results with media_url or thumbnail_cdn_url.
Ignore any result that lacks both renderable URL fields.
Choose the result that best matches the request rather than blindly taking the
first one.
This skill returns locators only: it does not download, upload, or turn an image into a Muse media reference. A Feed may use a suitable remote image URL. For an Artifact that needs a durable local copy, pass the chosen locator and source page through the Artifact's supported media-ingestion path; do not treat the CDN thumbnail as permanent storage.
For an Artifact, choose a download-safe locator before starting any fetch:
Inspect all returned results and prefer an HTTPS media_url with no URL
credentials, query string, or fragment. Preserve result order among those
simple public URLs.
Preflight each candidate like a browser, with a cross-origin referer, then fetch one candidate at a time with bounded connect and total timeouts (substitute real values for the placeholders):
curl -fsSLI -A 'Mozilla/5.0' -H 'Accept: image/png,image/jpeg,image/gif,image/webp,image/*;q=0.5' -H 'Referer: <any-https-origin-that-is-not-the-image-host>' '<image-url>'
curl -fsSL --connect-timeout 10 --max-time 60 -A 'Mozilla/5.0' -H 'Accept: image/png,image/jpeg,image/gif,image/webp,image/*;q=0.5' -o '<dest-file>' '<image-url>'
Accept only a final 2xx whose content type and file magic are an image,
then copy those bytes into Artifact-owned storage. Prefer PNG or JPEG:
WEBP is fine except in a Word document, and AVIF except on a web page,
which refuses it at share time with no retry that fixes it. Convert a HEIC
to JPEG anywhere. Reject hotlink-blocked, expiring, redirecting, 403/404,
non-image, watermarked, or unstable URLs. The referer matters: a
hotlink-protected host serves a plain fetch but refuses requests that look
like they come from someone else's page, so a URL that fails this probe
breaks after the Artifact ships even though it loads for you today. The
same preflight applies to any external image URL an Artifact uses, however
it was found.
If no simple media_url succeeds, consider a query-bearing media_url or
thumbnail_cdn_url. Use the returned locator byte-for-byte; never remove or
rewrite its query string to make it look simpler.
Search-result pre-approval records the returned locator's exact path and query, so a plain GET/HEAD that uses it byte-for-byte is normally allowed without another prompt even when it has a query string. Do not keep retrying or substitute an unrelated local image when a fetch fails.
This is text-to-image search, not reverse-image identification. Do not use it to identify an unknown person from a supplied photograph.
还没有评价。欢迎分享你的使用感受。
无需登录即可发表。欢迎分享使用感受、提问，或说明在哪些环境下无法运行。
还没有评论。来写第一条吧。
Create a 3D object with three.js, inspect it from every angle in an interactive viewer, and download it as OBJ + MTL or GLB for use in other 3D applications.
中文简介正在准备中。当前显示英语简介。
Maintain a private task list with deadlines, next steps, waiting parties, and blockers. Update progress and dependencies, merge duplicate items, and archive closed tasks so ongoing commitments stay clear and easy to follow.
中文简介正在准备中。当前显示英语简介。
Coordinate a task from a verifiable goal through completion. Handle work directly or split independent tasks across separate agent sessions, track progress, verify results, and arrange pull request merging when included in the goal.
中文简介正在准备中。当前显示英语简介。
Create animated HTML pages with smooth scene transitions and playback controls. An editable timeline lets users adjust scene lengths and speeds, with fixed landscape or portrait layouts prepared for video export.
中文简介正在准备中。当前显示英语简介。
Review synced Apple Health records, including steps, heart rate, sleep, and workouts. Retrieve daily summaries or detailed samples for a chosen period, and check sync status and missing data before interpreting the results.
中文简介正在准备中。当前显示英语简介。
Create and edit Word documents and templates, reorganize text, replace images, and work with tracked changes or comments. Apply proper headings and lists, then render every page to check its appearance before delivery.
中文简介正在准备中。当前显示英语简介。