How it works
Each image is shrunk to 1,024 pixels and searched on the web with Google Cloud Vision's web detection, which finds pages that show the same image, whole or cropped. The verdict comes from where those pages are, by the fixed rules below: we tested them on stock and news photos and on photos never posted online before launch. Visually similar images don't count; only the same image does.
The four verdicts
The image (or a crop of it) is on a paid stock site: Getty Images, iStock, Shutterstock, Rex, Alamy, Adobe Stock, Dreamstime, Depositphotos, 123RF, Pond5, Bigstock, age fotostock, EyeEm or Stocksy. This is checked first. Agencies also sell copies of public-domain archive photos, so an old photo found there may still be public domain: check where it first came from.
Not on a paid stock site, but on a free-license source: Wikimedia Commons, Unsplash, Pexels, Pixabay, Flickr, the Library of Congress, NASA, rawpixel or Public Domain Pictures. Free sources still have terms (credit, no endorsement, share-alike): check the page.
Neither of those, but the image is on pages across 10 or more different sites. Photos that travel this widely without a named owner are usually news or agency photos, so treat it as licensed.
Fewer than 10 sites and no stock or free source. This is what your own photos look like, but also images that simply aren't indexed yet. When the image is on 1 to 9 such sites, the result says "On a few sites, owner unknown" and lists them. It never means free to use.
Photos of people get their verdict and any stock or free-source pages, and how many pages carry the image, but no list of those pages: this isn't a people search. Evidence of where the image appears, not legal advice.
Scanning a site
Scan a site reads its robots.txt and sitemap (WordPress, Shopify, Squarespace and most site builders publish one), or follows its links from the home page when there's no sitemap. On each page it collects the images (including the largest size in a srcset and the page's sharing image) and skips icons, logos and images smaller than 200 pixels. Free scans take up to 20 images, Pro up to 1,000. The report lists every image with its verdict and the page it's on, worst first, and downloads as CSV. Our crawler is described on its own page.
In Claude
Claude can't search the web by image on its own. Add the connector (claude.ai, or the command below in Claude Code), sign in, then ask. Images attached to a chat don't reach connectors, so Claude gives you an upload link for files; addresses work directly.
claude mcp add --transport http trace-any-image https://traceanyimage.com/api/mcp- “Where is this image from? [image address]”
- “Is this photo a Getty or Shutterstock image? I want to use it on our blog.”
- “Find the original of this picture.”
- “Check the images on mysite.com for any I might need a license for.”
- “Watch mysite.com and tell me when new images need checking.”
Getting the image in
- A file from your device: JPEG, PNG, WebP, GIF, AVIF, HEIC, BMP or TIFF, up to 20 MB.
- An image's address (right-click the image, then Copy image address), or a page's address to trace that page's main image.
- X, Instagram, Facebook, TikTok, Threads and Pinterest links aren't fetched (their terms). Save the image and upload it instead.
- Uploads are deleted 24 hours after upload. Reports are kept 30 days.