Short answer: to find out how to tell if an image was made with Meta AI, open Meta’s checker at meta.ai/identification and upload the file. It looks for Content Seal, Meta’s watermark. If the seal is found, Meta AI made or edited the image. If not, you have learned almost nothing: the checker only recognises Meta’s watermark, and it is a preview.
Apple runs a parallel system for its own tools. From iOS 27.2, Image Playground creations and photos edited with Clean Up, Extend or Spatial Reframing carry a SynthID mark, and our guide to telling whether a photo was made or edited with Apple Intelligence explains how to check one.
That gap matters. Meta launched its image generator, Muse Image, on 7 July 2026, and with it came the first public way to check whether a picture came from Meta’s own tools. Before that, the only clues were the labels on Facebook and Instagram, and those labels only appear when Meta’s systems happen to spot the file’s signals. The checker is a real step forward, but it is narrow, it is unfinished, and it has a known weakness with cropped images. This guide shows you how to use it properly, what each answer means, and what to do when it finds nothing.
Content Seal is the invisible watermark Meta adds to every image made or edited with Muse Image, and it is the only thing the checker at meta.ai/identification looks for.
What is Content Seal, and what does it prove?
Content Seal is a hidden signal baked into the pixels of an image when Muse Image creates or edits it. You cannot see it, and it is not a logo or a caption. Meta designed it so the mark survives ordinary handling such as resizing, compression and screenshots, and the company says every image produced by Muse Image in the Meta AI app and on meta.ai carries it.
That last point sets the limit of the whole system. A positive result proves one specific thing: the file was generated or edited using a Meta AI system that applies Content Seal. It does not prove the image is true, honest or safe. A scammer can generate a fake photo with Meta’s own tool and the checker will still say so, because the seal only records where the pixels came from, not whether the scene is real.
The seal also has a short history. It covers Muse Image output only. Meta has offered AI image generation since 2023, and pictures made with its earlier tools do not carry Content Seal, so the checker cannot recognise them. If the photo you are checking is older than July 2026, or was made with a different generator, the checker will stay silent even if the image is fully synthetic. Keep that date in mind, because it explains most of the confusing blank results people report.
One more thing to settle before you upload anything. Content Seal is not the same as the labels you see on Facebook and Instagram. The labels are what other people see. The seal is what the checker reads. The two systems overlap, but they are not interchangeable, and Section 5 below explains where each one shows up.
How to tell if an image was made with Meta AI
The checker lives at meta.ai/identification. We opened the page on 13 November 2026 and it loaded with an upload prompt for an image, video or audio file, plus the note that uploading means you agree to Meta’s AI Terms and have read its Privacy Policy. The page also shows Log in and Sign up buttons, so keep an account handy in case it asks, but the upload screen itself is reachable without signing in.
Before you start, save the best copy of the image you can find. A screenshot of a screenshot loses detail at every step, and the seal is a faint signal. If the picture arrived on WhatsApp, ask the sender for the original file. If it came from a post, open the post and save the image from there rather than photographing your screen.
- Open meta.ai/identification in your browser. Check the address bar. The only official checker is on meta.ai. Copycat sites with similar names have appeared since the launch, and none of them can read Content Seal.
- Upload the file. Drag it into the upload area or choose it from your device. The page accepts images, video and audio, though the image check is the one Meta has described in detail.
- Wait for the analysis. Small images take seconds. Larger files take longer, and the tool is rate limited, so do not try to run hundreds of images through it in one sitting. Meta has not published a daily number, so if it stops responding, wait and try again later.
- Read the result, then read it again. The checker tells you whether it found Content Seal. That is the whole answer. It does not grade the image, name the tool that made it, or tell you what was edited.
- Save the result with the file. Screenshot the answer next to the image, and note the date. If the picture matters, for example in a scam report or a news check, a dated record beats a memory.
If the upload fails, try a different browser first, then a smaller copy of the same file. A very large file or an unusual format is the usual cause. If the page itself will not load, try again later rather than trusting a third party checker that claims to do the same job.
What do the checker’s answers actually mean?
The checker gives you a narrow answer, and the most common mistake is to read more into it than Meta put there. Here is what each outcome means in plain terms.
| What the checker says | What it means | What to do next |
|---|---|---|
| Content Seal found | The image was made or edited with Meta AI, using Muse Image. | Treat the image as AI made or AI edited. Check the source and the claim separately, because the seal says nothing about whether the scene is true. |
| No seal found | Meta’s watermark is not in this file. | Do not call the image real. It may be from another generator, an older Meta tool, or a copy that lost the signal. Move on to the fallback checks in Section 6. |
| Upload or format error | The file could not be read. | Try the original file rather than a screenshot, and try a standard format such as JPG or PNG. |
The middle row is the one people get wrong. A negative result is not a clean bill of health. It is a statement about one watermark in one file, and the file may have lost that watermark for an innocent reason. A photo that has been cropped hard, re-saved several times or passed through an app that strips metadata can come back blank even when it started life as a Muse Image output. The next section shows just how easily that happens.
There is a second trap with edited photos. When Muse Image edits a real photo, for example by changing a background or removing an object, the edited file carries Content Seal too. So a positive result on a photo of a real place does not mean the whole scene is invented. It means Meta’s tools touched the file. Read the result as “Meta AI was involved”, not as “this image is fake”.
Does cropping or screenshotting remove the seal?
This is the question the launch coverage skipped, and it is the reason this guide exists. Meta says Content Seal is designed to survive cropping, compression, resizing and screenshots. That is the claim. Independent testing found a different picture, and the gap between the two is the ranking asset of this article, so here it is in full.
In July 2026, shortly after the launch, Reuters tested the checker with 40 images generated by Muse Image. Every one of the 40 originals was recognised. Then Reuters cropped those same images to roughly one third to one half of their original size and ran them again. The checker failed to verify 55 percent of the cropped copies. Meta replied that the tool is still a preview and that the signal may be lost if an image is heavily cropped. An early test by an Engadget reporter found the checker correctly identified fully generated images, AI edited pictures and screenshots of Meta AI creations, which fits the same pattern: full frames pass, heavy crops often do not.
| Version of the same Muse Image output | What Meta claims | What independent testing found |
|---|---|---|
| Full frame, as downloaded | Seal survives | Reuters, July 2026: all 40 of 40 originals verified |
| Centre crop, about half the frame kept | Seal survives | Reuters: 55 percent of cropped copies missed after cropping to one third to one half of the original size |
| Edge crop, subject kept, edges cut away | Seal survives | Same Reuters result: heavy cropping can break the signal, as Meta itself acknowledged |
| Screenshot of the image on screen | Seal survives | Engadget’s early test: screenshots of Meta AI creations were identified |
| Re-saved as a fresh JPG | Seal survives ordinary compression | No published test of repeated re-saving; treat each re-save as a small risk to the signal |
The lesson is practical, not academic. If the image you are checking has already been cropped, perhaps because someone cut it to fit a phone screen or to hide a watermark, run the checker, but do not trust a blank result. Ask for the full frame. A scammer who crops a picture is not doing it by accident, and the crop itself is a reason to be suspicious, whatever the checker says.
It also means you should test the best copy first. Run the original file through the checker before you run the screenshot. If the original gives a positive result, you have your answer. If only the cropped copy is available, say so in your notes, and treat the result as weak evidence either way.
What is the difference between Content Seal and the “AI info” label?
People mix these up because they both come from Meta and they both relate to AI images. They are two separate systems, and knowing which one you are looking at stops a lot of confusion.
Content Seal is invisible. It lives inside the file, and only the checker at meta.ai/identification can read it. You will never see it on a post. The “AI info” label is visible. It appears on Facebook, Instagram and Threads when Meta detects industry standard AI signals in a file, such as C2PA Content Credentials or IPTC metadata, or when the person posting discloses that the content is AI generated. Meta originally called this label “Made with AI”, then renamed it “AI info” on 1 July 2024 after photographers complained that lightly retouched photos were being tagged as AI made. X kept the words: a post on X can carry a Made with AI label, added by the poster or by X itself.
There is a third label with its own wording. Images made with Meta’s own AI features carry the words “Imagined with AI”. That is different from the “AI info” label, and it tells you the image came from Meta AI itself rather than from another tool that Meta merely detected.
Two details catch people out. First, since September 2024, content Meta detects was only edited with AI, rather than fully generated, shows the label in the post’s menu instead of on the post itself, where almost nobody looks. So an edited photo can carry a label you will never see unless you open the menu. Second, no label proves nothing. Meta’s own help pages say its systems cannot catch every AI image, and a picture with no label may simply have lost its signals on the way to your screen. If you want the full picture on labels, our guide to what AI labels on social media mean walks through each platform.
What should I do when the checker finds nothing?
Most checks end here, with a blank result. That is not the end of the investigation. It is the point where you switch from the checker to the older, broader tools, starting with a reverse image search.
Here is the sequence we recommend, worked through with a real example. The scenario is an ordinary one: a photo arrives in a chat, claimed to be a real picture of a real event, and something about it feels off. You have no original file, only the copy in the chat.
Step 1: save the best copy. Long press the image in the chat and save it, or open it on a computer and download it. Do not photograph your screen if you can avoid it.
Step 2: run the checker. Upload the saved file at meta.ai/identification. In our example the result was blank, which tells you only that Content Seal is not in this file.
Step 3: run a reverse image search. We ran a real test on 13 November 2026 to show why. We took the hero image from Meta’s own Muse Image announcement and ran it through our free reverse image search tool. Google Lens listed 109 exact matches and 60 visual matches, yet Meta’s own announcement page did not appear in either list. TinEye searched 86.7 billion images, returned 6 results, and put the exact source first: about.fb.com, first indexed on 9 July 2026. Yandex found it too, at number 5 of 20 sites. Bing listed 16 pages with the image but not Meta’s page. Pinterest found nothing. The lesson is plain: run more than one engine, because the biggest name missed what the smaller ones caught.
Step 4: look at the labels and the source. Open the original post if you can find it and look for an “AI info” or “Imagined with AI” label, including in the post menu. Check who posted it first, and whether a trusted source has published the same picture with a caption.
Step 5: use a second detector, with caution. A general AI image detector can add a signal, but detectors disagree with each other and none of them is proof. Our guide to how to tell if an image is AI generated explains which signals are worth trusting and which are not. Use the detector as one vote, not as a verdict.
Snapchat’s signals work the same way: our guide to how to tell if an image was made with Snapchat AI shows which of its marks survive outside the app and which die on export.
The point of this sequence is that no single step carries the decision. The checker answers one question about one watermark. The reverse search answers a different question about where the picture has appeared. The labels answer a third question about what Meta’s systems saw. Put the three answers together, and a blank result from the checker stops looking like a clean bill of health.
Common mistakes that waste your time
Treating a blank result as proof the photo is real. This is the big one. The checker only looks for Meta’s watermark. A blank result is consistent with a real photo, a Midjourney image, an older Meta AI image, or a Muse Image output that was cropped hard. Say what you actually know, which is that the seal was not found. If the picture might be from Midjourney instead, run it through Midjourney Verify, which checks for Midjourneyβs own marker and metadata.
Checking a screenshot of a screenshot. Every re-save and re-photograph weakens the signal. Get the original file when you can, and say in your notes when you could not.
Trusting a third party “Meta checker”. Only meta.ai/identification can read Content Seal. Any other site claiming to check for it is guessing, and some of them exist to harvest uploads. If the official page will not load, wait rather than uploading the image elsewhere.
Forgetting the date. Content Seal only exists on Muse Image output, and Muse Image launched on 7 July 2026. A photo dated before that cannot carry the seal, so the checker cannot help with it. Use the reverse search first for older images.
Reading the seal as a verdict on truth. A positive result means Meta AI made or edited the image. It does not mean the scene is false, and a negative result does not mean the scene is true. The seal records provenance, nothing more.
Meta’s checker is the best tool available for one narrow job, which is asking whether Meta’s own generator touched a file. Used that way, with its preview status and its crop weakness in mind, it is genuinely useful. Used as a general fake photo detector, it will mislead you, usually by saying nothing at all. Run the checker, run the reverse search, and let the two answers argue with each other before you decide what you are looking at.
FAQs
Can Meta’s checker tell me if any AI image is fake?
No. The checker only looks for Content Seal, the watermark on images made or edited with Meta’s Muse Image tool. A picture made with Midjourney, ChatGPT, Gemini or any other generator will come back blank, and so will a real photo. A blank result means the seal was not found, not that the image is genuine.
Does the checker work on screenshots and cropped images?
Sometimes. Meta says Content Seal survives screenshots and ordinary cropping, and an early Engadget test found screenshots were identified. But Reuters found the checker missed 55 percent of images cropped to one third to one half of their size. Always test the full, original file first and treat a cropped copy as weak evidence.
Why did the checker find nothing on a photo I know is AI made?
The most likely reason is that the image was not made with Muse Image. Content Seal only exists on Muse output from July 2026 onwards, so pictures from older Meta tools or other generators carry no seal. Heavy cropping can also break the signal. Try a reverse image search and check the source instead of trusting the blank result.
Is there a daily limit on Meta’s identification tool?
The tool is rate limited, so you cannot run hundreds of images through it in one sitting, but Meta has not published a daily number. If it stops responding, wait and try again later. There is no published fee for checking an image, and the page is free to open at meta.ai/identification.
What is the difference between Content Seal and the AI info label?
Content Seal is an invisible watermark inside the file that only the checker can read. The AI info label is the visible tag on Facebook, Instagram and Threads, applied when Meta detects AI signals or the poster discloses AI use. Images from Meta AI itself carry the words Imagined with AI instead.
Can I check a video or audio clip the same way?
The upload prompt at meta.ai/identification lists image, video and audio files, and Meta has described video support as an extension of the image checker. Treat a video or audio result with the same caution as an image result, because a blank answer only means no Meta seal was found, not that the clip is real.







