How to Remove AI Info From an Instagram Post

Instagram AI info label: what it marks and where it appears on a post

The AI info label is a notice Instagram attaches to a post when the uploaded file carries data that says an AI tool made or edited the image.

Meta announced automatic labelling of AI images on 6 February 2024, saying it would read the AI generated markers "in the C2PA and IPTC technical standards" written by other companies' tools. The label reached a wider range of content from May 2024, under a different name

It launched as "Made with AI". Meta renamed it to "AI info" on 1 July 2024, after photographers found real photographs marked as AI creations for small retouching edits. Meta said the earlier labels "weren't always aligned with people's expectations and didn't always provide enough context", and that retouching tools were writing the "industry standard indicators that were then labeled 'Made with AI'".

Placement changed a few months after that. On 12 September 2024 Meta said the label would move for content only edited with AI tools, coming off the face of the post and into the post menu, with rollout the following week. Content detected as fully AI generated keeps the visible label.

Meta's own Transparency Center counted the label between 1 October and 29 October 2024. Instagram carried over 330 million labelled pieces in those 29 days, made up of nearly 140 million posts, over 220 million reels and over 24 million stories, and those carried over a trillion label views between them. This is not a rare label.

Where the AI info label shows on a feed post, a reel and a story

On a fully AI generated image the label sits under the account name on the post. On a lightly AI edited image it sits inside the three dot post menu instead.

Meta also said it would show whether a post is labelled "because of industry-shared signals or because someone self-disclosed". That line tells you which of the two paths applied to your post, and it is the quickest way to work out whether a file signal caused it.

Stories and reels follow the same rules. The Transparency Center figures above count all three surfaces separately, which confirms the label is not a feed only feature.

Does the AI info label reduce Instagram reach

Meta has never published a distribution penalty for a correct AI info label on a post.

The reach consequence Instagram has published applies to accounts rather than posts, and only when a label is missing. Instagram said on 31 August 2026 that it will "limit the reach of profiles that feature an AI-generated person and do not add this label". That sentence is about profiles, and no matching sentence exists for a labelled post.

Treat any specific percentage you read elsewhere as invented. Meta publishes label view counts and has published no post level reach figure.

Instagram AI info label triggers: the four signals read at upload

Instagram scans every uploaded image for four kinds of AI evidence, and any one of them can switch the label on without a human ever seeing the photo.

A metadata cleaner reaches three of these four signals. It cannot reach the fourth.

Instagram does not need to look at your picture to make the decision. I wrote a Photoshop style AI declaration into a 4032 x 3024 test JPG and then mapped the file. The string compositeWithTrainedAlgorithmicMedia sat at byte 943 of a 563,712 byte file, and pixel data did not start until byte 4,654, so the whole AI claim lives in the first 0.8 percent of the file, ahead of the image.

Adding the AI declaration to my clean file cost 3,539 bytes, about six tenths of one percent of the file size. A tag that small decides how the post is presented to every viewer.

Why Instagram labels a real photograph as AI

A real photograph gets the AI info label when an editing tool writes an AI provenance tag into the file, which happens with Generative Fill, AI denoise, AI masking, background removal and most AI upscalers.

What triggers the label is that tag, not the camera and not the picture. My test file carried ordinary iPhone EXIF and 49 exiftool tag lines before editing, and 54 lines after I added the editor's provenance fields. Those five new lines are the entire difference between a photo that reads as yours and one that reads as machine made.

This is the complaint that made Meta rename the label in July 2024, and the underlying behaviour did not change with the name.

What trainedAlgorithmicMedia means in your file

trainedAlgorithmicMedia is the IPTC Digital Source Type value that means the image was created by an AI model, and compositeWithTrainedAlgorithmicMedia means AI was used on part of a real image.

Both come from the IPTC Digital Source Type NewsCodes vocabulary, last revised 23 October 2024. The value travels as a URI rather than a word, so in the file it reads http://cv.iptc.org/newscodes/digitalsourcetype/trainedAlgorithmicMedia.

The second value is the one that catches photographers, because a real photo with an AI removed lamppost is not trainedAlgorithmicMedia. It is compositeWithTrainedAlgorithmicMedia, and it still carries an AI declaration.

To read the values in your own file, check what AI metadata your photo carries.

Does Instagram strip EXIF data from uploads

Instagram strips most EXIF data from the file other people can download, and it reads the AI signals first, so the stripping happens after the label decision is made.

Every upload gets re encoded to save bandwidth, and re encoding rebuilds the file from decoded pixels, so whatever the old container held goes with the old container. Meta describes the reading half of that cycle in its 6 February 2024 announcement, where it identifies the C2PA and IPTC markers at upload.

People assume that because Instagram deletes metadata, metadata cannot hurt them. Instagram reads, decides, labels, and only then strips.

AI info removal on a live Instagram post: the delete, clean and repost method

There is no button that removes the AI info label from a published Instagram post, so you delete the post, clean the original file, and upload it again.

1. Delete the labelled post, because the label is attached to that upload and cannot be edited off it.
2. Open the original file, not the copy you downloaded back from Instagram.
3. Strip the AI provenance data with a metadata cleaner, then confirm the tags are gone before you go near the app.
4. Upload the clean copy as a new post.

Step 3 is where the actual work happens, and you can remove AI metadata from an image in the browser without installing anything.

A clean pass on my AI tagged test file took it from 563,712 bytes and 54 exiftool tag lines down to 559,649 bytes and 19 tag lines. That is 4,063 bytes and 35 tag lines deleted, and every AI field came back empty afterwards, including the Digital Source Type, the CreatorTool and the Software tag.

Can you turn off the AI info label in Instagram settings

No setting in the Instagram app switches the label off, because the label is a response to what is inside your file rather than an account preference.

You can add the label yourself when you post AI content, and Meta counts that as self disclosure. You cannot subtract it. The only control you hold sits upstream, in the file and in your export settings.

How to remove the AI info label from an Instagram story or reel

The method is identical for stories and reels, with one difference: a reel cover image and a story background are separate files, and each one must be cleaned.

A cleaned video with a dirty cover still gets labelled, because the cover is uploaded as its own image and read as its own image. Clean the video, clean the cover, then check both before posting. If you built the cover in the same editor that triggered the label, assume it carries the same tag.

Does a screenshot remove the AI info label

A screenshot removes every metadata signal because it builds a new file from pixels only, and it also costs resolution and leaves any SynthID watermark intact.

I rebuilt my AI tagged file from its decoded pixels, the way a screenshot tool does. The PNG came out at 1,862,239 bytes, more than three times the size of the original, with 20 tag lines and not one provenance field. The JPG version came out at 557,554 bytes with 23 tag lines and the same empty result.

A real screenshot is worse than my test, because it captures at screen resolution rather than at the size of the photo, so a 4032 x 3024 photo comes back at whatever the phone display holds. A screenshot also does nothing to a pixel watermark.

How to appeal an Instagram AI label you believe is wrong

Post level AI info labels have no published appeal route, and account level restrictions are contested through Account Status in the app.

Instagram documents the Account Status path for the profile label, where creators can see reach limits and "appeal restrictions they believe were made in error". No equivalent form exists for a single labelled photo, so the practical remedy for a post stays the delete, clean and repost method above.

Keep the original camera file and the edit history either way. If the question ever escalates, an untouched original with intact camera EXIF is the evidence that carries weight.

AI info prevention at export: where Photoshop, Lightroom and Gemini write the tags

Turning off Content Credentials in your export settings prevents the AI info label at the source, which is faster than cleaning every file after the fact.

The last column has to carry your own date, because vendor behaviour changes between releases and an untested matrix will mislead the reader.

Two rows rest on vendor documentation rather than on a test. OpenAI states that supported images from ChatGPT, Codex and the OpenAI API carry both C2PA Content Credentials and SynthID, and it warns that those signals "can sometimes be removed by platforms, editing tools, or file conversions". Google announced on 20 November 2025 that images from Nano Banana Pro in the Gemini app, Vertex AI and Google Ads carry embedded C2PA metadata.

Two more rows record an absence. A blank vendor statement for Canva or CapCut means no published commitment was found, and it is not a test result. Fill those cells yourself.

How to turn off Content Credentials in Photoshop exports

Choose None in the Content Credentials section of the export dialog, at File > Export > Export As.

Adobe documents three options there. Publish to Content Credentials cloud stores them in Adobe's cloud and keeps the file smaller, Attach to file embeds them in the exported image, and None "doesn't include Content Credentials with this export". Content Credentials only apply to JPG and PNG exports.

The document level switch sits elsewhere. Window > Content Credentials (Beta) opens the panel with Enable Content Credentials, and the default sits at Photoshop > Settings > History & Content Credentials, which turns credentials on for new and saved documents. Turning that preference off stops the credential being built in the first place.

Lightroom keeps the same control under About Lightroom > Preferences > Export, where Apply Content Credentials governs whether they go out with the file.

AI info label limits: what metadata cleaning cannot reach

Cleaning metadata removes the signals stored in the file header, and it does not remove SynthID, which lives in the pixel values themselves.

SynthID is a Google DeepMind watermark that alters pixel values in a pattern a detector can read. Google DeepMind states it is "designed to stand up to modifications like cropping, adding filters, changing frame rates, or lossy compression", and Google reported more than 20 billion pieces of content watermarked with it by 20 November 2025.

I hashed the decoded pixels of my AI tagged file, stripped every metadata field out of it, and hashed the decoded pixels again. Both came back as 2a2c5e6911a3a477a3dd8eb5ee314c3a, so the strip deleted 4,063 bytes and did not change a single pixel.

A metadata cleaner works on the container and never on the image, so anything written into the image survives it by definition. If your file came from a Google model, no metadata tool will make it undetectable, whatever that tool's marketing says.

When the AI info label comes back after a repost

A label that returns on a cleaned file means the signal is in the pixels or in a copy you did not clean.

Three causes account for almost all of it: the reel cover was never cleaned, the app pulled a cloud synced original rather than your cleaned copy, or the file was re downloaded after cleaning and picked the tags back up.

Check which file the app actually uploaded before you clean anything a second time. A repeat label on a genuinely clean file points at the pixels, and that is not a cleaning problem.

AI generated profile label and AI info label: two different Instagram labels

The AI info label marks a single post based on the file you uploaded, and the AI generated profile label marks a whole profile based on whether the person shown in it is real.

Instagram introduced the profile label as the AI creator label and renamed it to "AI-generated profile" on 31 August 2026. Instagram's own wording explains the reason: people "don't like seeing a profile that seems human, only to find out later that the person featured is AI-generated".

The consequences here are published. Instagram will "limit the reach of profiles that feature an AI-generated person and do not add this label", and detected profiles without it stop appearing in recommendations. Two routes restore it: add the label, or appeal through Account Status. Instagram states that creators who add the label proactively "will not see a change to their profile's reach".

The profile label applies to accounts featuring an AI generated person, not to people who use AI tools in their work, so a photographer who runs AI denoise is not in scope for it.

Cleaning your image files does nothing about the profile label, because that label is about the subject of the account and not about the file.

AI disclosure law and the AI info label: where removal crosses a line

Removing an AI tag from a photograph you actually took is a correction, and removing it from a fully AI generated image you present as real is a disclosure failure that Article 50 of the EU AI Act now covers.

Article 50 transparency obligations took effect on 2 August 2026. The European Commission requires providers of generative AI systems to mark synthetic output, and the mark has to be "effective, reliable, robust and interoperable machine-readable". Deployers have to disclose deepfakes of real people. A grace period runs to 2 December 2026, and it applies only to generative AI systems placed on the market before 2 August 2026.

The obligation lands on providers and deployers rather than on every person with a photo app, but the line it draws still tells you which uses this page supports.

If the label describes your file accurately, removing it is a lie about the file. If it does not, removing it is a repair.

Final verdict on removing AI info from an Instagram post

Removing the AI info label works reliably when the trigger is a metadata tag written by an editing tool, and it fails when the image carries a pixel watermark, which makes prevention at export the only complete fix.

Delete, clean and repost handles the common case, because the common case is a real photograph carrying an editor's provenance tag. It cost 4,063 bytes to fix in my test file and left the picture untouched. It will not help a Google generated image, and nothing that works on metadata will.

Set your export settings once so the tag never gets written, then clean the file before the next upload.

Sources

• Meta, Labeling AI generated images on Facebook, Instagram and Threads, 6 February 2024, about.fb.com
• Meta Transparency Center, Labeling AI Content, updated 19 February 2025, transparency.meta.com

First hand file measurements taken 9 September 2026 with exiftool 12.76 on a 4032 x 3024 test JPG. Every primary source verified 9 September 2026. I did not attach images because you can use your own images on this tool and verify the results.