Reverse image search finds copies of the same picture: the identical photo, resized, cropped or recompressed, plus visually similar images and objects. Face search finds the same face in different pictures by comparing facial features. Use reverse image search to trace where a specific photo came from; use face search when you need other photos of the person in it, such as checking whether a match's photos belong to someone else.
The two are often confused because both start with an upload. This guide explains how each one works under the hood, what each can and can't find, and which to use for common situations, with a decision table at the end.
Three questions an image can answer
Before choosing a tool, decide which question you are asking:
- "Where has this exact photo appeared?" That is a copy question. Reverse image search answers it.
- "What is in this photo?" That is a recognition question: a product, a plant, a landmark, a dog breed. General visual search (Google Lens, Bing Visual Search) answers it.
- "Where else does this person appear, in any photo?" That is a face question. Only a face search engine answers it.
The questions overlap but answering one does not answer the others. A brand-new selfie may have zero copies online while the same person has hundreds of other photos indexed. A photo of a sunset has plenty of copies and no face at all.
How reverse image search works
Reverse image search engines describe a whole image in a compact form and look for other images with a close description in their index.
Fingerprints and features
Engines use different techniques, but the common ideas are:
- Perceptual fingerprints (hashes). The image is shrunk and simplified into a short code that stays similar when the image is resized, recompressed or slightly colour-adjusted. Two copies of the same photo produce near-identical codes.
- Local features. Distinctive points (corners, edges, textures) are detected and described. Matching enough of these points lets an engine recognise a crop, a rotated version or a photo with a logo pasted on.
- Learned visual embeddings. Modern engines also run the image through a neural network that captures what it looks like overall. This is how they return "visually similar" images and identify objects, even when no copy exists.
What comes back
- Exact and near-exact copies: the same photo on other pages, in different sizes and crops.
- Visually similar images: different photos with similar composition, colours or subject.
- Recognised entities: products, landmarks, artworks, text.
- Source pages: the web pages where each match was found, which is where the actual evidence lives.
Google describes Lens results as including matching objects, similar images and websites with the image. TinEye focuses on exact and modified copies and lets you sort them by date, which helps find the oldest appearance.
Where it fails
- A different photo of the same person usually won't match. To a copy engine, two selfies of one person taken a week apart are different images.
- Heavy edits (mirroring, strong filters, collages, large overlays) can push a copy outside the engine's tolerance.
- Unindexed pages: private accounts, content behind logins and very new posts.
How face search works
A face search engine ignores most of the picture and works only with faces.
The four steps
- Detection. The engine finds each face in the image and draws a box around it. If several faces are present, you may need to choose one or crop the photo first.
- Alignment. The face is rotated and scaled so the eyes, nose and mouth sit in standard positions. This reduces the effect of head tilt and framing.
- Embedding. A neural network trained on faces converts the aligned face into a list of numbers (a face embedding). The network is trained so that different photos of the same person produce nearby embeddings and photos of different people produce distant ones.
- Search. The engine compares your embedding against the embeddings of the faces in its index and returns the closest ones, usually with a similarity score and a link to the page each face came from.
Because it compares faces, not pictures, face search can find the same person with a different hairstyle, in another location, years apart, or in a group photo. It can also return people who simply look alike. How to read those results is covered in face search accuracy and similarity scores.
Where it fails
- No usable face: side profiles, sunglasses, masks, very small faces in a crowd.
- Index coverage: a face search only finds pages it has crawled and processed. Many people have little or no public photo presence.
- Lookalikes: similar-looking people can rank high, especially with low-quality input.
- Not built for objects: it won't identify a product, a place or a pet.
Which engines do face search
General engines such as Google Lens, Bing Visual Search and TinEye are designed around copies and objects. They don't offer a "find this person in other photos" feature. Our Google face search guide explains what Google does and doesn't do with faces. Dedicated face search engines include PimEyes and Facehunch, among others; the best face search engines page compares them.
Facehunch runs face searches for $4.99 per search, with no subscription. Every result shows its source link and similarity score. The uploaded photo is sent to our search provider for the search, and removal is requested afterwards.
Side-by-side comparison
| Reverse image search | Face search | |
|---|---|---|
| Compares | The whole image | Only the face |
| Finds | Same photo (resized, cropped, recompressed), similar images, objects | Same person in different photos |
| Good input | The original or largest version of the image | A clear, front-facing face, cropped to one person |
| Typical tools | Google Lens, Bing Visual Search, TinEye, Yandex | PimEyes, Facehunch |
| Typical cost | Free | Usually paid per search or by subscription |
| Main false positive | Visually similar but unrelated images | Lookalikes |
| Main blind spot | New photos of a known face | Images without a usable face |
| Best for | Tracing a photo's origin, finding reposts, stock photo checks, product and place identification | Checking whether profile photos belong to someone else, finding your own face online |
Which one to use: decision table
| Your situation | Start with | Then |
|---|---|---|
| A dating match's photos look professional or too good | Reverse image search on each photo | Face search if nothing turns up |
| A match's photos look like casual selfies | Face search | Reverse search to find the original posts |
| You suspect a profile picture is a stock photo | Reverse image search (TinEye, sort by oldest) | Usually nothing more is needed |
| You want to know where a news or viral photo first appeared | Reverse image search | Check captions and dates on the oldest results |
| You want to find out if someone is using your photos | Reverse search for your known photos | Face search to catch photos of you that you didn't know were online |
| You want to identify a product, plant or landmark | Google Lens or Bing Visual Search | Not a face search task |
| You think a photo might be AI-generated | Reverse image search for earlier versions | See AI image and deepfake checks |
| You saw a stranger and want to know who they are | Neither | Face search is not for identifying strangers |
Two rules follow from this table:
- Don't dismiss a concern using the wrong tool. "Reverse image search found nothing" says nothing about whether the face appears elsewhere. "Face search found nothing" says nothing about whether the exact photo is a stock image without a clear face.
- Use both when the stakes justify it. For checking a dating profile, a free copy search takes two minutes; a face search adds coverage for photos taken from private people's social accounts. The dating profile catfish guide walks through the full sequence.
A practical workflow
- Write down your question in one sentence. "Are these photos of the person I'm talking to?" is a different job from "Where was this photo first published?"
- Keep the original file untouched and make a working copy. Crop the working copy to what matters: the face for face search, the full image or the key object for copy search. When searching from an app screenshot, crop out buttons, names and text overlays.
- Pick the method using the table above.
- Open the source pages. Thumbnails show resemblance; pages show who published the image, when, and with what caption.
- Record what each page supports, including anything that conflicts. A photo on a stock site shows it is available there; it doesn't show who controls an account using it.
- Stop when your question is answered. Don't keep searching for more details about a person once you have what you need.
Why both methods miss things
No engine sees the whole internet. Each one has an index: a selected collection of pages and images it has crawled and processed, which is always incomplete and somewhat out of date. Private social accounts, messaging apps, content behind logins and newly posted photos are mostly invisible to all of them.
So an empty result always means "not found in this index at this time", not "doesn't exist online". If a search returns nothing, try the other method, try a different photo, and read reverse image search not working for technical causes. Scores from different engines also can't be compared, because each engine defines its own scale.
Privacy and responsible use
Uploading a photo to any search engine sends it to that provider. Read the provider's terms on retention before uploading, especially for photos of other people. Search a crop of the face rather than a full photo with other people, documents or locations in the background.
Face search is for checking someone who has already contacted you, protecting yourself from fraud, or finding where your own face appears. Don't use it to identify strangers, find where someone lives or contact someone who doesn't want contact. If you find your own photos misused, the find my photos online page explains the next steps.
Written by Facehunch with AI assistance and checked against the cited sources. Our editorial policy.