Remove the AI label from photos and videos before you post
How Instagram, Facebook, Threads, TikTok and YouTube decide to label a post as AI, what inside the file tells them, and how to post a clean copy of a photo or video.
4 min read
On this page7 sections
Short answer. Automatic AI labels come from data inside the file, not from how the picture looks. Instagram, Facebook, Threads and TikTok read C2PA Content Credentials and AI metadata fields at upload and label the post if they find them. Remove that metadata, re-save the file, and the automatic label does not appear. HideSlop does it for images and videos, and the first file is free. Labels you add yourself, and labels a reviewer adds, are a different matter.
Where the label comes from
Every AI generator and most AI editing features leave a trace in the file. Platforms look for four of them:
| Signal | Where it lives | Who writes it |
|---|---|---|
| C2PA Content Credentials | A signed manifest in the JPEG, PNG, WebP, MP4 or MOV container | ChatGPT, Gemini, Firefly, Photoshop, Microsoft Designer, Samsung Galaxy AI, Sora, Veo |
| IPTC digital source type | An XMP field that says "trained algorithmic media" | Midjourney, Google Magic Editor, many generators |
| Tool names in EXIF, XMP or QuickTime metadata | Software, CreatorTool and encoder fields | Almost every generator and editor |
| AIGC labels | An XMP block required by China's AI content rules | Kling, Hailuo, Doubao, Jimeng and other Chinese tools |
The first two are what Meta and TikTok act on. The others identify the tool to anyone who looks, and a file with no camera data at all reads as edited.
Platform by platform
| Platform | Automatic label | Reads | Self-disclosure |
|---|---|---|---|
| Instagram, Facebook, Threads | "AI info" | C2PA, IPTC | AI label toggle; required for realistic AI |
| TikTok | "AI-generated" | C2PA | AI-generated content toggle; required for realistic AI |
| YouTube | Shows how a video was made when credentials exist | C2PA | Altered or synthetic content disclosure |
| A CR icon on the image | C2PA | None | |
| "AI modified" | Metadata and its own classifiers | None |
Two things follow. A file without Content Credentials or an IPTC AI tag gets no automatic label anywhere. And the platforms do not need the label to know the tool: a reviewer can still see "ChatGPT" in the metadata.
How to remove the AI label from an image
- Open the image remover and drop the image in, or paste it from your clipboard. JPG, PNG and WebP work, up to 20 at once.
- Press Remove AI content. The C2PA manifest and every AI field in EXIF, IPTC and XMP are removed. The image is re-saved as a JPEG at phone quality (92 to 97 percent) with camera metadata from a real iPhone model, and the pixels get an invisible variation: a one to five percent zoom, a slight color shift, a little grain.
- Save to Photos on iPhone or download the ZIP on desktop, then post.
The output is a new file. Platforms that match uploads against earlier ones (reposts, carousels reused across accounts) see it as a fresh photo.
How to remove the AI label from a video
- Open the video remover and upload an MP4, MOV or WebM up to 150 MB.
- Press Remove AI content. C2PA and AI tags come out of the container, and the video is re-encoded with a two to eight percent zoom, a speed change of a couple of percent, a small timing offset, light grain and fresh camera metadata. Nobody watching can tell.
- Download it or send it to your phone, then post it as a new upload.
Most videos finish in under a minute; long ones take a few. Every run uses fresh random values, so a second version has a different fingerprint from the first.
Check before you post
The free AI label checker shows what a platform will read: Content Credentials and who signed them, the IPTC AI field, AIGC labels, generator names and camera data. Run the original to see why you were labeled, and run the clean copy to confirm nothing is left. It works on your device, with no upload.
What this does not do
- It does not remove labels you added. Switch the platform's AI toggle off when you post the clean file.
- It does not remove visible watermarks. A logo drawn into the picture (Gemini's sparkle, Sora's moving mark, Meta AI's "Imagined with AI") is pixels, not metadata.
- It does not remove SynthID. Google's invisible watermark lives in the pixels and is built to survive edits. HideSlop changes every pixel slightly, which disrupts simple fingerprinting, but it makes no claim about SynthID. The honest state of play is in can you remove SynthID?
- It does not change platform rules. Realistic AI content still has to be disclosed where the platform requires it.
FAQ
How do I remove the AI label on Facebook?
Facebook uses the same system as Instagram. Remove the C2PA and IPTC metadata from the image or video, then post the clean file. A label already on a published post cannot be switched off.
Can I remove the "Made with AI" label?
"Made with AI" was renamed "AI info" in 2024. The trigger is unchanged, so the fix is the same: a clean file, posted again.
Does a screenshot remove the AI label?
Usually, because a screenshot has no Content Credentials. It also throws away resolution and leaves the file with no camera metadata, which looks edited. A proper re-save keeps the quality.
Will the platform still know it is AI?
Automatic labels rely on metadata, and that is gone. A human reviewer, a visible watermark, or an obviously synthetic scene can still lead to a label, and the platform's disclosure rules still apply to you.
Which generators does this work for?
All of them. The remover rewrites the whole file instead of hunting for a known tag, so ChatGPT, Gemini, Midjourney, Firefly, Stable Diffusion, Flux, Sora, Veo, Runway and Kling outputs are handled the same way.