Free, private, no guessing

AI Image Checker

See what an image says about how it was made: Content Credentials, AI labels and the prompt, checked in your browser.

Last updated: November 2026

C2PA

Signatures verified

IPTC

AI labels

Prompts

SD, ComfyUI, NovelAI

Private

No uploads

Drop your image here

JPG, PNG, WebP, HEIC, AVIF, TIFF and more. One image or many.

Your file never leaves this device

How to check if an image says it was made with AI

01

Add the original file

Drop or choose the image, ideally as you first got it: downloaded from the AI tool, saved from the camera or sent as a file. Screenshots and copies saved from social media have usually lost what this checker reads.

02

Read the verdict

The checker says what the file claims about how it was made, such as "Says it was made with AI", and how far that claim can be relied on: a confirmed signature, an unconfirmed one, or just a label anyone could write.

03

Look at the evidence

Open Content Credentials to see who signed them and each step, from the newest back to the first. Prompt and settings shows what an image generator saved, with a Copy button.

04

Check another way if nothing turns up

No AI information doesn't mean no AI. See other ways to check below, such as Google's SynthID check in Gemini.

What the checker reads

AI tools, cameras and editing apps can save information about an image inside the file, next to the picture. This checker reads all of it that concerns AI, and nothing else: it never guesses from how the picture looks.

Content Credentials (C2PA)

A signed record of how the image was made and edited. Checked with the official C2PA library against the official trust list, including earlier steps stored inside it.

IPTC AI labels

The Digital Source Type field, such as "Created using Generative AI", plus the AI system, version and prompt fields photo metadata can hold.

Prompts and settings

What Stable Diffusion WebUI, ComfyUI, InvokeAI, NovelAI, Fooocus and SwarmUI save in their images: prompt, negative prompt, model, seed, steps and more.

Hidden pixel data

NovelAI also hides its generation details in the lowest bits of the transparency channel. The checker reads them even when the text was stripped.

Tool names

Software and creator fields that name an AI tool, such as Adobe Firefly. Shown as a hint, since any app can write them.

What it doesn't do

It doesn't judge the picture, guess from how it looks, or read invisible watermarks such as SynthID.

Standards: C2PA specification, IPTC Digital Source Type vocabulary.

What the results mean

Says it was made with AI

The file states the image was created with generative AI: Content Credentials or an IPTC label say "Created using Generative AI", or an image generator saved its prompt and settings.

Says it was edited with AI

The file states generative AI changed or added part of the image, such as inpainting a photo, or that it's a composite with AI elements.

Says it was taken with a camera

Content Credentials or an IPTC label say it was captured from real life, by a camera or a scanner, with no AI step recorded.

Mentions an AI tool

Only a tool name, in a field like Software. A hint worth knowing, not a statement that the picture was generated.

No AI information found

The file says nothing about AI. It may be a photo, or an AI image whose information was removed or never added.

How far to rely on it

  • Signature confirmed

    Content Credentials signed by a company on the official C2PA trust list, unchanged since.

  • Signature not confirmed

    Signed, but the signer isn't on the trust list or the certificate doesn't meet the rules. The statement may still be right; who made it can't be confirmed.

  • Changed after signing

    The image or its credentials changed after they were signed, so they no longer match.

  • The file's own claim

    An IPTC label or generator data. Useful, but anyone can write or remove it.

Even confirmed Content Credentials show who made a statement, not whether the picture is true. As the Content Authenticity Initiative puts it, "Content Credentials don't indicate if an image is fake." Source: Content Authenticity Initiative FAQ.

What AI tools save in their images

We checked images straight from ChatGPT and Gemini, and read the source code of the image generators people run themselves. Here is what each one leaves in the file:

  • ChatGPT

    Content Credentials: created with ChatGPT (gpt-image), converted, invisible watermark added. PNG files, tested November 2026.

  • Google Gemini

    Content Credentials: "Created by Google Generative AI" and "Applied imperceptible SynthID watermark", sometimes inside an earlier step. JPG files, tested November 2026.

  • Stable Diffusion WebUI and Forge

    Prompt, negative prompt and settings in a PNG text field, or in the EXIF comment of JPG and WebP files.

  • ComfyUI

    The prompt graph and the full editor workflow, in PNG text fields, or in the EXIF of WebP files.

  • InvokeAI

    Its generation details as JSON in a PNG text field.

  • NovelAI

    Prompt and settings in PNG text fields and hidden in the transparency channel.

  • Fooocus and SwarmUI

    Their settings as JSON in a PNG text field.

Other tools that follow the C2PA or IPTC standards are read the same way. ChatGPT and Gemini don't save the prompt in their images, so there is none to show.

Can AI metadata be removed?

Yes. Content Credentials, IPTC labels and prompts sit next to the picture in the file, and they are easy to lose. A screenshot starts a new file without them: "A screenshot of an image wouldn't include C2PA metadata from the original image." Re-saving in many editors drops them too, and the Content Authenticity Initiative notes that "it is still common for social media and content platforms to remove metadata."

That is why a missing label proves nothing, and why this checker never says an image is real. It is also why AI companies add invisible watermarks, such as Google's SynthID, which are part of the pixels rather than the metadata. Durable Content Credentials pair the two: a watermark carries an identifier that can be used to look up the record again after the metadata was stripped.

We don't offer a tool to remove AI labels. Hiding that an image was made with AI misleads the people who see it, and removing the metadata doesn't remove an invisible watermark.

Sources: Content Authenticity Initiative FAQ; The three pillars of provenance (Content Authenticity Initiative, January 2026).

Other ways to check an image

Google SynthID, in Gemini

On gemini.google.com or in the Gemini app on Android, upload the image and ask "Is this AI-generated?". Gemini looks for Google's SynthID watermark, which is part of the pixels rather than the metadata, but it only recognizes content made with Google's AI tools, and Google lists a limit of about 10 image checks a day. A found watermark means all or part of the image was made or edited by Google AI; none found doesn't rule out other AI. Source: Gemini Apps Help, Verify AI-generated images.

The official Content Credentials Verify site

The Verify site from the Content Authenticity Initiative shows the same signed records. Unlike this checker, it uploads your image.

Find where the image first appeared

A reverse image search, such as Google Lens, Bing Visual Search or TinEye, often leads to the original post or a larger copy, which may say how it was made or still carry its Content Credentials.

Detectors that judge the picture

Some tools estimate from the pixels whether an image looks AI-made. They give a likelihood, not an answer, and can be wrong both ways, so treat a result as one clue among others.

Private by design

Your images are read in your browser and never uploaded. When an image has Content Credentials, the checker loads the official C2PA library from this site, about 3 MB, once, and checks the signature against a copy of the official C2PA trust list bundled with the site. It doesn't fetch credentials stored online or look up certificates on the internet, so nothing about your image leaves your device.

Built with the C2PA JavaScript library (MIT) and the C2PA trust list (CC BY 4.0), which is copyright the Coalition for Content Provenance and Authenticity.

Common mistakes when checking images

Checking a screenshot or a saved social media copy

Both usually lose Content Credentials and metadata. Check the original file, from the AI tool, camera or sender.

Reading "No AI information" as "real"

It only means the file says nothing. Many AI images lose their labels on the way to you.

Treating any label as proof

IPTC labels and generator data can be written by anyone. Only signed Content Credentials show who made the statement.

Trusting camera details

A camera make and model in EXIF can be copied into any image, AI images included.

Why use ImgTweak?

✓No upload, no account
✓Official C2PA library and trust list
✓Shows every step, not just the last
✓Prompts and ComfyUI workflows
✓Many images at once
✓Works on phones

Frequently asked questions

Can this tell if any image is AI generated?

No tool can tell that for every image. This checker reads what the file says about itself: Content Credentials, AI labels and generator settings. When those are there, it tells you what they say and whether their signature holds up. When they've been removed, or were never added, it can't tell, and it says so instead of guessing.

What are Content Credentials?

Content Credentials are a signed record inside an image of how it was made and changed, following the open C2PA standard. AI tools such as ChatGPT and Gemini add them to images they make. The signature shows who made the record and whether the image changed afterward.

What does "Signature confirmed" mean?

The Content Credentials were signed by a company on the official C2PA trust list, and neither the image nor the credentials changed after signing. It confirms who made the statement, not that the statement is true about the picture.

Why does a ChatGPT image show "Signature not confirmed"?

In our tests some ChatGPT images were signed with a certificate that doesn't meet the C2PA rules, so the official library can't confirm the signature, though nothing says the image changed. The credentials still say the image was made with ChatGPT; the checker shows that, with the reason the signature fails.

Why does it say "No AI information found" for an AI image?

The information was probably removed. Screenshots don't keep it, and many apps and sites remove metadata when you upload or save an image. Some generators never add any. Try the original file, or check another way, such as Google's SynthID check in Gemini for images made with Google AI.

Can it find the prompt used to make an image?

Yes, when the image generator saved it. Stable Diffusion WebUI, ComfyUI, InvokeAI, NovelAI, Fooocus and SwarmUI save the prompt and settings in the file, and some tools add an IPTC prompt field. ChatGPT and Gemini don't save the prompt in their images.

Can it read ComfyUI workflows?

Yes. When a ComfyUI image holds its workflow, the checker shows the prompt, model and sampler settings and offers the workflow as a JSON file you can load back into ComfyUI.

Does it detect SynthID or other invisible watermarks?

No. Invisible watermarks are hidden in the pixels and only their makers' detectors can read them. When an image's Content Credentials say a watermark was added, the checker tells you. For Google's SynthID, upload the image to Gemini and ask whether it was made with Google AI.

Is camera information proof that a photo is real?

No. Camera make and model are plain text that any app can write or copy. Content Credentials signed by a camera or phone carry more weight, because their signature breaks if the image is changed.

Which file types can it check?

JPG, PNG, WebP, HEIC, AVIF, TIFF, GIF and JPEG XL. A file the browser can't preview, such as HEIC in some browsers, is still checked.

Are my images uploaded?

No. Everything runs in your browser. The official C2PA library and its trust list load from this site only when an image has Content Credentials, and they send nothing anywhere.

Can AI labels be removed from an image?

Yes, metadata labels and Content Credentials can be removed, often by accident when an image is re-saved or posted. That's why their absence proves nothing. Invisible watermarks are harder to remove, and durable Content Credentials use them to find a removed record again. We don't offer a tool to remove AI labels.

Ready to check an image?

Add an image above to see what it says about how it was made.