To reverse image search a screenshot, crop it to the picture itself, removing status bars, buttons, captions and chat bubbles, then search that crop. If the results are poor, search a second, wider crop that keeps the background. The same rule covers most difficult images: give the engine the real picture with nothing added, and never "fix" it with AI enhancement, because invented detail is not evidence.
Search engines compare what they are given. A screenshot that is half app interface will match other app screenshots. A mirrored, sideways, compressed or text-covered photo can fall below the match threshold. This guide shows how to prepare each kind of awkward image, and what makes a good photo for a face search.
The two-copy rule
Before changing anything, make two files:
- The original, exactly as you received or captured it. This is your record. Do not edit it.
- A search copy that you crop, rotate or convert.
Every later step works on a fresh copy made from the original, never on a copy of a copy. Each re-save of a JPEG loses a little more detail, and editing the original makes it impossible to show later what you actually saw.
A short log helps too: which version you searched, on which engine, and what it found. "Tight crop, Google Lens: stock photo site, 2019" takes ten seconds to write and saves a lot of repeated searching.
Quick reference: what to do with each kind of image
| Image problem | What to do | What not to do |
|---|---|---|
| Screenshot with app interface | Crop to the photo only | Search the whole screen |
| Already tightly cropped | Search it, then any wider version you can find | Use AI "expand" to invent edges |
| Possibly mirrored | Search the original and a horizontally flipped copy | Assume selfies are always flipped |
| Sideways or upside down | Rotate and export a corrected copy | Confuse rotating with flipping |
| Blurry or heavily compressed | Find a less compressed source; search as is | Sharpen, upscale or "restore" faces |
| Inside a PDF | Export or screenshot just the illustration; note page and caption | Upload a confidential document |
| Black and white | Search as is; compare structure | Colourise it |
| Transparent PNG cut-out | Search as is, then on a plain background | Add a decorative background |
| Scanned or photographed print | Straight, glare-free capture, cropped to the print | Generative restoration |
| Frame or border | Search with and without the frame | Crop off a signature or credit |
| WebP file | Upload directly if accepted; otherwise export JPEG or PNG | Rename the extension |
| Text or meme overlay | Search the full image, a clean region, and the text | Use AI text removal |
The sections below explain each one.
Screenshots: crop out the interface
Screenshots are the most common input, and the easiest to improve.
- Open the screenshot in Photos, Gallery or your desktop image viewer.
- Use the crop tool to keep only the picture: no status bar, no username, no like button, no chat text.
- Keep enough of the background to be useful. A distinctive wall, view or object often leads to the original post.
- Save the crop as a new file and search it.
- If results are unrelated, search a wider crop that includes more of the scene.
Cropping also protects other people. A screenshot of a chat or a group can include names, messages and faces that have nothing to do with your search. Remove them before sending the image to any external service.
If you can save the actual image instead (long-press and Save, or right-click and Save Image As), do that. It is almost always sharper than a screenshot of it. Device-specific steps are in the guides for iPhone and Android and Mac and Windows. For a still from a video, see how to search a video frame.
Cropped photos: tight crop vs wide crop
Different questions need different crops.
- Tracing where a photo was published: keep the whole composition, including background, clothing and objects. These features are what link a copy to its source.
- Identifying an object or product: isolate it so a bigger item in the frame does not dominate the results.
- Face search: one face, filling a reasonable part of the frame (see the face photo section below).
Imagine a portrait with a distinctive orange chair behind the subject. Cropping away the chair makes the face clearer but removes the feature that connects it to the original magazine page. Search one tight and one wide version and compare.
Do not use generative "expand" or "uncrop" tools to fill in missing edges. The new pixels are an AI guess. They can produce matches that look convincing and mean nothing.
Mirrored and flipped images
People who reuse photos sometimes flip them horizontally, either by habit or to make copies harder to find. Selfie cameras can also save a mirrored version.
Clues that an image has been flipped:
- Text, logos or numbers read backwards.
- A parting, mole, scar or jacket badge on the "wrong" side compared with other photos.
- Watches, wedding rings or shirt buttons on the unexpected side.
To test it, duplicate the image in an editor, apply only a horizontal flip to the copy (in Photos on iPhone: Edit, Crop, then the flip icon; on a Mac in Preview: Tools, Flip Horizontal) and search both versions. Some engines handle flipped copies well, others do not, so testing both is cheap insurance.
A flip on its own does not prove bad intent. It is often a normal editing choice. Record what you found and where, and let the context speak.
Rotated and sideways images
Phones store orientation as a metadata tag rather than physically rotating the pixels. Some apps read that tag, others ignore it, which is why a photo can look upright in your library and sideways in an upload preview.
- Open a copy in a trusted editor.
- Rotate it until it is upright.
- Export or save it, so the rotation is written into the image itself.
- Check the exported copy in a second viewer before uploading.
Rotating (turning the image) is different from flipping (mirroring it). Note which one you did.
Compressed, blurry and low-resolution images
Images forwarded through messaging apps, re-shared on social media or saved as thumbnails lose detail every time. You will see blocky squares around edges and smeared fine textures.
- Find a better source first. The original download beats a screenshot of the download, which beats a forwarded copy. If someone sent you the image, ask for the original file.
- Search the image as it is. Engines often cope with modest compression.
- Do not sharpen or upscale. Sharpening exaggerates the compression blocks. AI upscalers and face restorers invent eyes, teeth and skin texture that were never in the photo, which makes face matches unreliable.
- Lean on surviving features. Composition, clothing and distinctive objects survive compression better than fine detail.
- Try text search for readable text. If a sign or caption is legible, search the words.
If nothing survives, label any match as tentative and keep the low-quality original with your notes.
Pictures inside a PDF
Searching a whole PDF page sends a pile of unrelated text and layout. Isolate the picture.
- Note the page number, caption and any photo credit. These are often more useful than the image.
- Use your PDF viewer's export option if it has one, or zoom in and take a screenshot of just the illustration.
- Search the image, and separately search a distinctive phrase from the caption.
- Compare candidate sources with the PDF's own references.
Do not upload a confidential document to search one public illustration in it. A match to the same chart or photo does not confirm the report's surrounding claims, and if the PDF gives no credit, describe the image as unattributed rather than crediting the first result you see.
Black-and-white photos
A monochrome copy keeps shapes, shadows and layout but loses colour. Search it as it is; most engines still find copies of the same photograph.
Do not colourise it first. AI colours are guesses, and they change the signal the engine compares. When you find a likely colour version, compare structure: the same fold in a sleeve, the same background geometry, the same shadow angles. A colour version is not automatically the original. Publishers often produce both.
Transparent PNGs and cut-outs
A cut-out (a person, product or logo with no background) can display differently in each tool: some show transparency as black, some as white, some as a checkerboard.
- Search the transparent PNG directly if the service accepts PNG.
- If the preview looks wrong, export a copy on a plain white or grey background and search that.
- If you have the full image the cut-out came from, search it as well. The complete advert or photo usually has more distinctive features.
Finding the same cut-out on several design sites does not tell you who created it or whether it is licensed for reuse. Check the source listing's terms.
Scanned and printed photos
For an old print, a careful capture matters more than anything else.
- Use a flatbed scanner if you have one. Otherwise photograph the print with the phone held parallel to it, in even daylight, with no flash.
- Avoid glare and reflections, especially on glossy prints or photos behind glass.
- Crop to the picture's edges, removing the tabletop or album page.
- Note any studio stamp, handwriting or caption on the back and search those words separately.
Many family prints were never published online, so an empty search is normal and says nothing about the people in them. Do not use generative restoration to "fill in" faded faces.
Frames and decorative borders
Decide what you are looking for:
- The artwork or photo: crop the frame off and search the image only.
- The framed product as sold: keep the frame, because the listing photo will show it.
Search both if unsure. Never crop off a signature, watermark or publisher's mark: that is the attribution you are trying to find. If you photographed a framed print at an angle, a simple perspective straightening is fine; keep it as a labelled search copy.
WebP files
WebP is a modern web image format. Facehunch, Google Lens and many other engines accept it directly, so try the original before converting. If a site rejects it, open the file in an image editor and export a JPEG (or PNG if it has transparency).
Check that the download really is an image. A browser's "Save Page As" can create an HTML file even when you were looking at a picture. Right-click the image and use Save Image As instead.
Text, memes and overlays
When text covers part of a photo, you have three separate clues.
- Search the complete graphic. Memes and quote cards are often reposted whole.
- Crop the largest clean region of the underlying photo and search that.
- Search a distinctive phrase from the overlay as plain text.
The text may lead to whoever made the graphic, while the picture leads to a stock library or the original photographer. Those are different sources with different rights. Do not use AI "remove text" tools; the filled-in area is invented, not recovered.
What makes a good photo for face search
Face search compares facial features, not the whole picture, so the ideal input is different from a general reverse image search.
| Good | Poor |
|---|---|
| One person in the frame | Group photos (crop to one face) |
| Face roughly front-on, eyes visible | Strong profile, face turned away |
| Even light | Hard shadows, backlit silhouettes |
| Face reasonably large in the image | Tiny face in a wide shot |
| No sunglasses, masks or heavy filters | Beauty filters, face reshaping, AI enhancement |
| The original or a clean screenshot | A thumbnail forwarded several times |
If you have several photos, try the clearest one first. Facehunch accepts JPG, PNG or WebP up to 10 MB and works best with one clear face. It costs $4.99 per search, shows the source link and a similarity score for every result, sends the photo to its search provider with removal requested afterwards. A better photo improves the odds of a useful match, but a high score is still a lead to check, not proof of identity; the face search accuracy guide explains how to read scores.
Only search faces you have a good reason to check: your own, someone you are already talking to, or a profile you suspect is using stolen photos. Do not use face search to identify strangers.
Duplicate photo finders are a different tool
If you are trying to clean up copies in your own photo library, you do not need a web search at all. The Photos app on iPhone and Mac has a Duplicates album under Utilities that groups identical and near-identical photos and lets you merge them. Uploading a private library to a web engine will not help with this and exposes your photos needlessly.
Web reverse image search answers a different question: where does this picture appear online? Use it when you want to find reuse of your images or a better-quality original. If you find your photos being used without permission, see how to remove images from search results.
Written by Facehunch with AI assistance and checked against the cited sources. Our editorial policy.