See what your image says about itself before you post it.
Platforms use more than file data to decide on an AI label, including their own detection. This tool reads only the markers stored in the file, so it shows what your file says, not whether a platform will label it. Nothing is uploaded, and your file is not changed.
Drop or click to choose
Drag & drop or paste from clipboard
JPEG, PNG, WebP, HEIC, AVIF: 25 MB max
100% privateNo sign-upReads only
Where an AI label comes from
A label can come from three places, and only the first is something you can see in the file:
Metadata. Information stored in the file, such as C2PA Content Credentials and the IPTC digital source type. AI tools write it, and so do some editing apps after a generative edit.
An invisible watermark. A signal in the pixels, such as Google's SynthID, which Google says is designed to stand up to cropping, filters and lossy compression.
The platform's own checks. YouTube says it may label content that its internal systems detect, and any platform can act on your own disclosure or a manual review.
What each platform says it does
Platform
What triggers a label
What you have to disclose
Instagram, Facebook, Threads
C2PA and IPTC AI signals, or your own disclosure. AI-edited content shows the label in the post's menu, AI-generated content on the post.
Realistic AI video and audio. Meta says it may apply penalties if you do not.
YouTube
Your disclosure, its own AI tools, C2PA metadata, its internal detection or a manual review.
AI-made or altered content that seems realistic. Fantasy content, minor edits and production help such as a thumbnail or script are exempt. Repeatedly skipping it can bring a label, removal or suspension from the Partner Program.
TikTok
Content Credentials on uploads from certain other platforms, or your own label.
Realistic AI content.
These rules change. Check the platform's help page before relying on any of this.
What the tool looks for
C2PA Content Credentials: the record of how the file was made, and the app that wrote it.
IPTC digital source type: a standard value in the file's XMP. trainedAlgorithmicMedia means made by AI, and compositeWithTrainedAlgorithmicMedia means partly made by AI.
Tool names and settings: creator tool fields, plus the prompt and sampler text some image generators save.
If your file has a marker you did not expect, look at the app that last saved it. If you are making a thumbnail with AI art, our editor is free to use, and you can read YouTube's disclosure rules to see what applies to the video.
Why does Instagram or Facebook say my photo has AI info?
Meta says it adds the label when it detects industry-standard AI signals in a file, such as C2PA or IPTC data, or when someone discloses AI use. Drop your original file above to see whether it carries one of those markers.
Can a real photo get an AI label?
Yes. A real photo can carry an AI marker after one generative edit, because the marker records the edit. Meta renamed its label from "Made with AI" to "AI info" in July 2024 after people objected to it appearing on photos with small AI retouches. When Meta detects that AI only edited a post, the label now sits in the post's menu.
Does saving an AI image again remove its label?
Not reliably. A new file, including one saved from a browser canvas such as our editor, usually leaves the old metadata behind, so a checker may then find nothing. An invisible watermark is in the pixels and is designed to survive edits and compression. Platforms can also use their own detection, and a label already on a post stays there. The rules do not change either: realistic AI content still has to be disclosed where a platform requires it, whatever the file contains.
Can this tool remove an AI label?
No. It only reads your file and never changes it. A label on a post belongs to the platform, so it can only be changed there, and only where the platform allows it. YouTube says creators can change most automatic labels, but not those from C2PA metadata, from its own AI tools or from a manual review.
Do AI thumbnails need a label on YouTube?
YouTube lists using generative AI to help with a thumbnail, title, script or outline as production help that does not need disclosure. The video is a separate question: realistic AI-generated or meaningfully altered video does need it.
What is the difference between AI metadata and an invisible watermark?
Metadata is information stored alongside the picture inside the file, so a tool can read it and many exports drop it. An invisible watermark is woven into the pixels themselves, so it can survive a new file. This page reads the first kind only.
Does "no AI marker found" mean the image is not AI?
No. Screenshots, exports and platform uploads often lose metadata, and many generators never write it. All the result tells you is what this file says about itself.
Can this detect SynthID?
No. Only Google's own tools can check for it. Google says you can upload an image to Gemini and ask, or use its SynthID Detector.
Does it prove who made the image?
No. It reports what the file claims. It does not verify the cryptographic signature on Content Credentials, and metadata can be written by anyone, so treat a result as a lead.
Is my image uploaded?
No. The file is read in your browser with JavaScript and nothing is sent to a server. Files up to 25 MB are accepted: JPEG, PNG, WebP, HEIC and AVIF.