You spent hours crafting an image with AI tools, edited it further in Photoshop, and posted it to Instagram — only to see a "Made with AI" label slapped on it. Sound familiar? Here's exactly how platform detection works and what you can do about it.
How does Instagram detect AI images?
Instagram (and its parent company Meta) uses two primary methods to detect AI-generated content:
- C2PA metadata reading — Instagram reads the Content Credentials (C2PA/JUMBF) embedded in image files. If the signature chain traces back to an AI generator like DALL-E, Midjourney, or Firefly, the "Made with AI" label is applied.
- IPTC DigitalSourceType — A metadata field that indicates how the image was created. When set to "trainedAlgorithmicMedia" or similar values, Instagram flags it.
Instagram does not currently use visual AI detection (analyzing what the image looks like). The detection is entirely metadata-based.
Which platforms show AI labels?
- Instagram — "Made with AI" label on posts with C2PA from AI generators
- Facebook — Same system as Instagram (both Meta platforms)
- LinkedIn — "CR" badge for Content Credentials on images
- Pinterest — AI content flagging in certain regions
- Google Search — Reads IPTC data for image attribution in Search results
How to avoid the "Made with AI" label
Since platform detection is metadata-based, removing the metadata before uploading prevents the label from appearing:
- Use okie.fun's AI Strip to remove C2PA, EXIF, and IPTC metadata
- Select "Combined" mode for extra protection (also disrupts potential classifier detection)
- Use the AI Detector to verify the image is clean before posting
The entire process takes seconds and runs locally in your browser.
Does re-saving or screenshotting remove AI labels?
Re-saving as JPEG removes some metadata but not always C2PA. Screenshots strip all metadata but significantly reduce image quality. A dedicated metadata stripper like okie.fun is more reliable and preserves full image quality.
Will platforms eventually use visual detection?
It's possible. Research into visual AI detection (analyzing textures, noise patterns, and statistical anomalies) is advancing rapidly. However, these methods still have high false-positive rates and are not deployed at scale by major platforms as of 2026. okie.fun's "Combined" mode with LSB perturbation provides an additional layer of protection against potential future classifier-based detection.