Does Instagram detect AI images?
Meta reads C2PA and IPTC signals and asks you to disclose the rest. Every Meta rule below links to Meta's own page.
Partly. Meta reads industry signals, C2PA and IPTC metadata plus invisible watermarks, that AI tools write into files, and adds an "AI info" label when it finds them. Files without those signals go through unlabeled unless the creator discloses. Meta says it cannot catch all AI content, and it requires creators to disclose photorealistic AI video and realistic-sounding audio, with penalties for skipping it.
Make your first photosInstagram detects AI images only when the file carries a signal Meta reads: C2PA or IPTC metadata written by the tool that made it, or an invisible watermark from a company that applies one. When it finds one, it adds an "AI info" label. When it finds none, the post goes up unlabeled, which is why Meta also requires creators to disclose realistic AI content themselves. The standard behind most of those signals is published by the C2PA, the coalition behind Content Credentials, and if you run an AI model account the disclosure rule sits right next to the growth work in growing an AI model on Instagram from zero.
The rest of this page lists each rule Meta has published, with the date and a link to Meta's own page, then shows what that means for a creator posting AI images every day.
Proof
We did not run a detection test. This is a table of what Meta has published, taken from Meta's newsroom and transparency pages, each linked. Where a page gives a date, the date is in the table.
| Date | What Meta said | Source |
|---|---|---|
| 6 February 2024 | Meta puts visible markers, invisible watermarks and IPTC metadata on photorealistic images made with Meta AI | Meta newsroom |
| 6 February 2024 | Meta is building tools to detect the "AI generated" fields in the C2PA and IPTC standards, to label images from Google, OpenAI, Microsoft, Adobe, Midjourney and Shutterstock as those companies add the metadata | Meta newsroom |
| 6 February 2024 | Meta says it is not yet possible to identify all AI content and that people can strip invisible markers | Meta newsroom |
| 6 February 2024 | Creators must use the disclosure tool for organic posts with photorealistic video or realistic-sounding audio made or altered with AI; Meta may apply penalties if they do not | Meta newsroom |
| April 2024 | Labels on a wider range of video, audio and image content start in May 2024, based on industry signals or self-disclosure | Meta newsroom |
| 1 July 2024 | "Made with AI" renamed "AI info", because small edits such as retouching triggered the old label | Meta newsroom, update |
| 12 September 2024 | Content Meta detects as only edited with AI gets the label in the post menu; content detected as AI generated keeps the label on the post | Meta newsroom, update |
| Current page | Meta adds AI info labels when it detects industry-standard AI indicators or when people disclose AI content | Meta Transparency Center |
Meta reads machine signals where the tool that made the file wrote them, and it leans on creators for everything else.
How detection works
Meta's system does not look at a photo and judge whether it seems fake. It checks the file for signals that the AI tool put there.
The first kind is metadata. C2PA, also called Content Credentials, is a signed record attached to the file that says which tool made it. IPTC is an older photo metadata standard with a field called digital source type, and one of its values means "created using generative AI". When either field says AI, Meta labels the post.
The second kind is an invisible watermark: a pattern hidden in the pixels. Meta writes one into images made with Meta AI, and some other companies write their own.
Because both kinds come from the tool, detection depends on the tool. A file from a generator that writes no signal, or a file whose metadata was removed by a re-save, a screenshot or an editor, gives Meta nothing to read. Meta says this openly in its February 2024 post. That gap is the reason the disclosure rule exists.
What you must disclose
Meta's rule names two things outright: photorealistic video and realistic-sounding audio that were made or altered with AI. Meta says penalties can follow when a creator skips it.
For an AI model account the practical reading is simple. If a post shows a person who looks real and that person was generated, disclose it. That covers the reels you make in the AI video generator and, as our own advice, the photorealistic stills too. Meta can also add a more prominent label to content with a high risk of misleading people on a matter of public importance, so keep AI faces away from news, politics and real public figures.
If you schedule posts through PixelPig, the studio handles one part of this for you. The Content calendar chapter of the Academy shows the flow: connect a Business or Creator Instagram account, open a finished image in the Gallery and schedule it. Our posting code sends an AI flag with every Instagram post (isAiGenerated: true in supabase/functions/social-post/index.ts). Meta decides how that flag is shown, so confirm the AI info label appears on the live post.
Method: posting AI images the way Meta asks
- Make the image in the AI image generator from your trained persona.
- Decide if the post is realistic. A real-looking person is. A clear cartoon or illustration is not covered by the photorealistic rule.
- Disclose before you publish, either with Meta's AI label in the posting flow or through the PixelPig scheduler, which sends an AI flag; check the live post shows the label.
- Say it in the profile too. A short line such as "AI model" in the bio costs nothing and matches what followers see on the label.
- Keep the account on the right side of Meta's other rules. Labels do not make banned content allowed. Explicit material stays off Instagram, as the Instagram chapter of the Academy and the scheduling chapter both say.
Label outcomes compared
| Situation | Does Meta label it? | What the creator should do |
|---|---|---|
| Image from a tool that writes C2PA or IPTC AI fields, metadata intact | Yes, from the signal | Nothing extra, the label is already there; disclosing also does no harm |
| Image made with Meta AI | Yes, Meta adds its own markers | Nothing extra |
| No signal found in the file | No automatic label | Disclose it yourself |
| AI image from a tool that writes no signal | No | Disclose it yourself |
| Real photo touched up with an AI editor | Label sits in the post menu since September 2024 | Follow the edit rules for your use |
| Photorealistic AI video or realistic AI voice | Required disclosure | Use the disclosure tool, or penalties can apply |
Common mistakes
Treating "no label" as approval. A missing label only means Meta found no signal. The disclosure duty still applies, and the penalties in Meta's posts are for creators who skip it.
Assuming detection reads the picture. It reads the file. Two identical-looking images can get different results because one kept its metadata and one did not.
Disclosing in the caption only. A caption line helps followers, but Meta's own label comes from its disclosure tool or from detected signals. Use the tool.
Confusing edited and generated. Since September 2024 Meta treats them differently. A generated persona is generated, even if you later retouch it.
Forgetting the other platforms. TikTok has its own label and its own wording; the rules are in TikTok's rules for AI-generated content. If you build the account from scratch, creating an AI influencer step by step covers where disclosure fits in the setup.
Bottom line
Instagram detects AI images when the file tells it to: C2PA or IPTC metadata, or an invisible watermark. It does not promise to catch the rest, so Meta puts the job on creators, with penalties for skipping realistic AI content. Disclose every realistic AI post, keep the bio honest, and if you post through PixelPig, check that the AI flag the scheduler sends shows as a label on the live post.
Questions
- Can Instagram tell if a photo is AI generated?
- Instagram can tell when the file carries an industry signal it reads, such as C2PA or IPTC metadata or an invisible watermark from a tool that applies one. Meta says it cannot yet identify all AI content, so a file without signals can pass with no label.
- What does the AI info label mean on Instagram?
- It means Meta detected AI signals in the post or the creator disclosed AI use. Tapping it shows more detail. Since September 2024, content that Meta detects was only edited with AI shows the label in the post menu instead of on the post.
- Why did Instagram change Made with AI to AI info?
- Meta said in its July 2024 update that the old label did not always match what people expected, because small edits such as retouching could trigger it. The new name is AI info.
- Do I have to label AI images on Instagram?
- Meta requires disclosure for photorealistic video and realistic-sounding audio made or altered with AI. Our advice goes further: if a post shows a realistic AI person, disclose it, still or video.
- Does the AI info label reduce reach on Instagram?
- Meta has not published any statement that the label changes distribution. We found no official page saying it does, so we make no claim either way.
- Can you remove the AI info label from an Instagram post?
- Meta has not published a way to remove a label it applied from detected signals or your own disclosure. If a real photo was labeled only because of a small AI edit, Meta's September 2024 change already moved that label into the post menu.
- Does Instagram read EXIF data to find AI images?
- Meta names C2PA and IPTC metadata and invisible watermarks as the signals it reads. It does not name camera EXIF fields as an AI signal in any post we found.