AI Image Detector: Is This Image AI?
Find out if an image was made with AI. The detector reads the Content Credentials, AI labels and generator data inside the file and explains its verdict, without uploading your image.
Drop a PNG, JPG, WebP, AVIF or HEIC file, or paste with Ctrl+V. Use the original file, not a screenshot.
Checking…
fact_check Why we think so
data_object Metadata found in the file
No detector is certain. This check reads the AI labels, Content Credentials and generator data stored in the file. Screenshots, messaging apps and social networks often remove that data, and metadata can be edited, so "no AI markers" does not prove an image is real. Also look for a visible AI badge or sparkle in a corner, and for odd hands, text and reflections.
How to use AI Image Detector
-
1
Get the original file. Download the image itself rather than taking a screenshot. Screenshots and many social networks remove the metadata the detector reads.
-
2
Upload, drop or paste the image. Drop a JPG, PNG, WebP, AVIF or HEIC file into the upload area, browse for it, or paste it with Ctrl+V.
-
3
Read the verdict and the score. The result card says AI-generated, AI-edited, possibly AI, likely a real photo, or no AI markers found, with an AI likelihood meter and a confidence level.
-
4
Check each reason. Every signal is listed with its strength (strong, medium or weak) and where in the file it was found, so you can judge the evidence yourself.
-
5
Verify Content Credentials. If the file has a C2PA manifest, upload it to contentcredentials.org/verify as well to confirm who cryptographically signed it.
By Mohit Khatri · Last updated
Learn more about AI Image Detector
Direct Answer: The most reliable way is to read the provenance data inside the file. Most major AI image generators now attach a label saying the image was made by AI, either as C2PA Content Credentials (a signed record of how the file was made) or as the IPTC Digital Source Type value trainedAlgorithmicMedia in its XMP metadata. Local tools such as Stable Diffusion and ComfyUI save the full prompt and settings in the PNG file. If those markers are present, the image almost certainly came from an AI tool.
When the markers are missing, you have to rely on weaker clues: the image size, visible badges or watermarks in a corner, and visual mistakes such as malformed hands, garbled text, inconsistent shadows and reflections. None of these is proof on its own.
Tip: always test the original file. A screenshot, a WhatsApp forward or a re-upload to most social networks removes metadata, and the detector can then only report "no AI markers found".
| Signal | Where it lives | Who writes it | Strength |
|---|---|---|---|
| C2PA Content Credentials | JPEG APP11, PNG caBX, WebP C2PA chunk | OpenAI, Adobe Firefly, Google, Microsoft, Meta, some cameras | Strong |
| IPTC Digital Source Type | XMP metadata | Google Gemini and Imagen, Midjourney, Adobe, Shutterstock | Strong |
| Generation parameters | PNG text chunks (parameters, prompt, workflow) | Stable Diffusion WebUI, Forge, ComfyUI, InvokeAI, NovelAI | Strong |
| Generator name | EXIF Software, XMP CreatorTool, Credit, history | Many apps and services | Strong |
| Generator image size | Pixel dimensions (e.g. 1024 × 1024, 1792 × 1024) | Default output sizes | Weak |
| Camera EXIF | Make, Model, exposure, GPS | Cameras and phones | Counter-evidence |
- AI-generated: the file carries at least one strong marker, such as an AI label, Content Credentials from an AI tool or embedded generation settings.
- AI-edited: the labels say a real image was changed with AI, for example with generative fill (
compositeWithTrainedAlgorithmicMedia). - Possibly AI-generated: only medium hints were found, such as a generator name in a free-text description.
- Likely a real photo: the file has camera details and no AI markers. EXIF can be copied, so this is a hint, not proof.
- No AI markers found: nothing points either way, usually because the metadata was removed. This is inconclusive.
Direct Answer: metadata can be removed or written by hand, and pixel-based classifiers make mistakes on new generators, heavy compression and edited photos. Treat any detector result as one piece of evidence.
- Stripped files: screenshots, messaging apps and most social networks discard EXIF, XMP and Content Credentials.
- Invisible watermarks: some providers, such as Google with SynthID, also hide a watermark in the pixels. Those can only be checked with the provider's own tools, which this browser tool cannot run.
- Forged metadata: anyone can write a camera model into EXIF. Signed Content Credentials are harder to fake, which is why you should verify them at contentcredentials.org.
For an important decision, combine this check with a reverse image search, the source's reputation and a close look at the details.
- C2PA Technical Specification — the open standard behind Content Credentials.
- IPTC Digital Source Type vocabulary — the official values for AI-generated and AI-edited media.
- Content Credentials Verify — check the signature of a C2PA manifest.