Why Instagram labeled your real photo "AI info" — and how to prevent it

You photographed it yourself. You edited it in Lightroom, exported, posted — and Instagram slapped an "AI info" label on it. You are not alone, and your photo didn't get misjudged by some AI-detection algorithm scanning your pixels.

It got flagged by metadata. Which means you can see the trigger before you post, and decide what to do about it.

What the label is

Meta launched its AI label in May 2024 as "Made with AI," then renamed it to the softer "AI info" just two months later after a wave of photographers complained that their real photos were being tagged. The label now appears on Instagram, Facebook, and Threads when Meta believes an image was generated or edited with AI.

The key word is believes. Meta's primary signal isn't visual analysis — it's the provenance metadata embedded in your file: C2PA Content Credentials and IPTC digital-source tags that editing and generation tools write automatically. Upload a file carrying those marks and the label applies. Meta then strips the credentials from the file it serves, so viewers see the label but not the underlying data.

Why real photos get flagged

Here's the part that catches photographers: you don't need to generate an image with AI to end up with AI metadata in it. Adobe's tools embed Content Credentials describing what operations touched the file, and several common operations count as AI:

• Generative Fill / Generative Expand in Photoshop — even a tiny patch on a dust spot writes a Firefly-signed credential into the export
• AI Denoise and similar features in Lightroom — noise reduction that runs through an AI model can mark the file
• Object removal tools — content-aware and AI-powered removal features in various editors
• Sky replacement and one-click AI edits in editors that have adopted C2PA

The manifest doesn't distinguish between "generated this entire image from a prompt" and "removed a power line with generative fill." Both can read as AI involvement, and platforms err toward labeling. That's how a photo you took with a camera, of a real place, ends up labeled alongside Midjourney renders. Photographers have been filing complaints about exactly this since 2024, and the forum threads are still growing in 2026.

How to check your file before you post

The metadata that triggers the label is readable before you upload. Drop the exported file into the metadata viewer and look at the report:

• A C2PA / Content Credentials manifest — this is the primary trigger. The report shows who signed it (Adobe, OpenAI, etc.) and what it asserts.
• IPTC DigitalSourceType — values indicating synthetic or composite media are read by Meta and other platforms.
• XMP fields naming AI tools as the creator tool.

If none of those are present, Instagram has no metadata signal to act on. If they are, you know exactly what the platform will see.

Your three options

Option 1: Post as-is and keep the label. If the image genuinely is AI-generated or heavily AI-edited, the honest move is to leave the provenance intact. Meta's policies (and, since August 2026, EU transparency rules for deepfake-style content) expect meaningful AI content to be disclosed. The label isn't a punishment; it's context.

Option 2: Avoid the trigger at the source. If the flag comes from an incidental edit — a generative-fill dust spot on an otherwise real photo — you can re-export without the AI-touched step, or use a non-AI tool for that operation (the classic clone stamp doesn't write AI credentials).

Option 3: Clean the metadata before uploading. Strip the C2PA manifest and AI-related tags from the export, and the metadata trigger is gone. Our C2PA remover does this in your browser — the file never leaves your device — and shows you exactly which fields it's removing, including a heads-up when you're about to remove an authenticity signature.

We'd add one honest note on option 3: it removes the metadata signal, not every signal. Meta has said it also works on classifier-based detection, and pixel-level watermarks like Google's SynthID survive metadata removal entirely (more on that here). Cleaning metadata reliably prevents the false-positive case — a real photo flagged for an incidental AI edit. It is not a cloak for passing off fully synthetic images as real, and we don't recommend using it as one.

Why this problem is growing

Since August 2, 2026, EU law requires AI image generators to embed machine-readable marks in their outputs, with full coverage of existing tools required by December 2026 (our explainer). More tools embedding more provenance metadata means more files carrying triggers — and more incidental flags on lightly-edited real photos. Platforms, under the same regulatory pressure, will keep reading those marks aggressively.

The result is a widening gap between what the label implies to viewers ("this is an AI image") and what actually triggered it ("Lightroom's denoise model touched this file"). Until platforms get more granular about degrees of AI involvement — which Meta has acknowledged is the hard problem — the label will keep flattening that distinction.

TL;DR

Instagram's "AI info" label is driven by metadata: C2PA Content Credentials and IPTC tags that tools like Photoshop and Lightroom embed automatically, sometimes for edits as small as a generative-fill dust spot. Check your file before posting and you'll know whether it will be flagged. Keep the label if the AI involvement is real; re-edit or clean the metadata if it isn't. The one thing you shouldn't do is be surprised by it — the trigger is sitting in your file, readable, right now.