Guides

Short answers to specific questions about checking an image: what Content Credentials, metadata, invisible watermarks and platform labels can show, and what they cannot. Each guide says what to look for in the checker's result.

Checking an image in practice

• Adobe Firefly's Content Credentials: what they say, and why a checker may call them a legacy matchWhy an Adobe Firefly file can read as a legacy trust-list match: two trust lists, two answers. What Firefly says it records and what each list says.
• Do Gemini and Nano Banana images carry SynthID, a sparkle or Content Credentials?What Google says about SynthID and Content Credentials on Gemini and Nano Banana images, what our checker can read, and where to check the watermark.
• Does Canva add Content Credentials or AI metadata to its images? What two Canva files and Canva's own pages showWhat Canva's AI Product Terms say about C2PA metadata, and what our checker read in two files filed as made with Canva.
• Does ChatGPT add a watermark or metadata to its images?What OpenAI says ChatGPT images carry, what we found in real files, and why some OpenAI-named Content Credentials read as signer trust not established.
• Does Midjourney add a watermark or metadata to its images?Midjourney says every image it delivers has a hidden ID tag that its own checker reads. What the tag is, how easily it is lost, and what we find.
• How to check whether a photo has been editedNo tool can say a photo was never edited, but some signals show that it was. How signed records, metadata and re-saves behave, with test files.
• How to verify a photo someone sent you: a checklistA short checklist for a photo you did not take: ask for the original, read its signals, search for earlier copies and weigh the source. Why each step.
• Is this product photo AI? How to check listing and marketplace imagesHow to check a product or listing photo for Content Credentials and AI metadata, why a signed claim in a file is only a claim, and what a clean result means.
• Is this profile picture AI? Checking dating and social profile photosHow to check a dating or social profile picture for provenance signals, why shared photos lose them, and what finding nothing means. No face guessing.
• Microsoft Image Creator, Paint and Designer: what Content Credentials they add and how a checker reads themMicrosoft says Image Creator and Paint add Content Credentials. Why our checker reads two files as a legacy trust-list match and what the current list says.
• Reverse image search to verify a picture: Google Lens, TinEye, Bing and Yandex, step by stepHow to use Google Lens, TinEye, Bing and Yandex to see where else a picture appears, with the link behaviour we tested, and what a search cannot tell you.
• Stable Diffusion's invisible watermark: what it is, how to check for it, and when it survivesHow Stable Diffusion's reference software adds an invisible watermark, what survives a re-save, and the false-match test we ran on 250 photographs.
• Which AI image generators add Content Credentials, metadata or a watermark?A dated table of what ChatGPT, Gemini, Firefly, Midjourney and Microsoft say they add to an image file, next to what our checker read in 32 test files.
• Why AI image detectors disagree, and what a percentage score meansAI image detectors score pixels, and they often disagree. Why that happens, what a percentage means, and what reading a file's signals can add.

Content Credentials and C2PA

• Cameras and phones that sign photos with Content CredentialsWhich cameras and phones sign photos with Content Credentials, who is on the trust lists, and what a camera's signed claim does and does not prove.
• Can Content Credentials be faked? Forged certificates and what a checker must showCan Content Credentials be faked? A self-made certificate can carry any name. What tests on forged and tampered files show, and what a checker must show.
• Content Credentials versus IPTC and XMP metadata: a signed record versus a labelContent Credentials are a signed record; IPTC and XMP metadata are labels anyone can edit. How the two differ, with test files that show it.
• Durable Content Credentials: how a watermark and a fingerprint recover a stripped recordHow durable Content Credentials combine signed metadata, a watermark and a fingerprint to recover a record after it is stripped, and where that stops.
• How to add Content Credentials to your own images, and what a self-signed certificate shows in a checkerHow people add Content Credentials to their own images with open tools, and why a self-signed certificate reads as signer trust not established in a checker.
• How to check Content Credentials on an image, step by step, and what each result meansA step by step guide to checking Content Credentials in your browser, and what each result headline means, from trusted to signer trust not established.
• How to read a Content Credentials record: claim generator, actions, ingredients, signatureHow to read a Content Credentials record: claim generator, actions, ingredients, signature and source type, with what we saw in real ChatGPT files.
• The "Content credentials label added" notice on LinkedIn: what it means and what it does notWhat LinkedIn's Content Credentials icon means, what its credential panel shows, and why the label alone does not show trust in the signer or truth.
• The C2PA Trust List and the Interim Trust List: who is on which, and why it changes resultsThe C2PA Trust List and the frozen Interim Trust List: who is on each, how their dates differ, and why one file can read differently on the two lists.
• What "signer not trusted" or "signer trust not established" means in a C2PA checkWhat signer trust not established means in a C2PA check, why an older record naming OpenAI and a forged one read the same, and what a reader can still learn.
• Which platforms read or show Content Credentials? LinkedIn, TikTok, YouTube, Meta and GoogleWhich platforms read or show Content Credentials, what LinkedIn, TikTok, YouTube, Meta and Google say, and why a platform label is not a trust check.
• Why Content Credentials disappear: screenshots, messaging apps, social uploads and conversionScreenshots, messaging apps, social uploads and file conversion can drop a Content Credentials record. What vendors say, what we measured, and what a missing record means.

Invisible watermarks

• Can AI watermarks be removed? What survives cropping, screenshots and re-saving, and what the rules sayWhat survives cropping, screenshots and re-saving for invisible AI watermarks, what vendors and the law say about removal, with our test results. No tools.
• Does OpenAI use SynthID? What OpenAI Verify checks, and what it cannotOpenAI says its images carry SynthID alongside Content Credentials. What openai.com/verify checks, what it cannot, and what our checker reads.
• Does TikTok watermark AI content? Its invisible mark, Content Credentials and the AI-generated labelTikTok says it is testing invisible watermarks and reads Content Credentials. What that means, what TikTok's own checker covers, and how its label works.
• Invisible AI watermarks compared: SynthID, Meta's watermark, Adobe TrustMark and Stable Diffusion's markSynthID, Meta's watermark, Adobe TrustMark and Stable Diffusion's mark compared: who can read each, what each maker says, and which one our checker decodes.
• Meta's invisible watermark: what Meta checks on Facebook, Instagram and ThreadsWhat Meta says about its invisible watermark and AI labels on Facebook, Instagram and Threads, the limits it admits, and what our checker can read.
• SynthID for video and audio: what exists, where to check, and why image checkers do not cover itGoogle says SynthID covers video and audio as well as images. What exists, where to check, the quotas, and why an image checker like ours does not cover them.
• Watermark, metadata or fingerprint: three ways to tie a claim to an imageA watermark, a metadata label and a fingerprint each tie a claim to an image differently. What each survives, with one photo saved four ways as a test.

Metadata

• AI parameters in PNG files: the Stable Diffusion "parameters" chunk, ComfyUI workflows, and what they proveStable Diffusion front ends can write a prompt and settings into PNG text. What the parameters chunk holds, what ComfyUI adds, and what it proves.
• Can EXIF data be trusted? Dates, camera models and GPS can be editedDates, camera models and GPS in EXIF can be edited with ordinary software. Why EXIF is a label, not proof, and how a signed record differs from it.
• Does a screenshot keep metadata or Content Credentials?Does a screenshot keep metadata or Content Credentials? What vendors say, what a screenshot removes, and how to check the original file instead.
• Does sharing an image remove its metadata? What each route doesDoes sharing an image remove its metadata? What vendors say about stripping, why routes differ, and what we measured when we re-saved real files.
• How to view an image's metadata on iPhone, Android, Windows and Mac without uploading itHow to see an image's metadata on iPhone, Android, Windows and Mac without uploading it anywhere, and which fields matter when you check for AI claims.
• IPTC Digital Source Type values explained: what trainedAlgorithmicMedia and the others meanWhat the IPTC Digital Source Type values mean, from trainedAlgorithmicMedia to digitalCapture, with the values we found in real AI files and how to read each one.
• Where AI declarations live in a file: XMP, EXIF, IPTC, PNG text and C2PA, mappedWhere an AI declaration can sit inside a file: C2PA, XMP, IPTC, EXIF and PNG text, mapped, with examples from the real AI image files we tested.

Platforms, labels and law

• California's AI Transparency Act: latent disclosures, the required verification tool and the datesCalifornia's AI Transparency Act as amended by SB 1000 and AB 2713: hidden disclosures in AI images, the free verification tool and the dates that follow.
• China's AI content labeling measures: explicit and implicit labels since September 1, 2025China's labeling measures for AI-generated content: explicit and implicit labels, the metadata duty and the September 1, 2025 start, from the text.
• EU AI Act Article 50: what must be marked and disclosed from August 2, 2026, and the Commission guidelines of July 20, 2026What EU AI Act Article 50 says about marking and disclosing AI-generated images, the Commission guidelines of July 20, 2026, and what a check shows.
• Google Merchant Center and AI images: the IPTC Digital Source Type requirementGoogle Merchant Center asks for the IPTC digital source type on AI-generated product images. What to keep, what to add to titles, and how to check a file.
• Google's SynthID checks in Gemini, Search and Chrome: how to use themHow to ask Google whether an image carries SynthID in Gemini, Search or Chrome: quotas, file limits, and why a no result is not proof of anything.
• How TikTok applies its AI-generated label, and what the label provesHow TikTok's AI-generated label is applied, by creators, by Content Credentials or by TikTok's own systems, and what the label does and does not prove.
• Meta's "AI info" label on Facebook, Instagram and Threads: what triggers it and what it provesWhat Meta's AI info label on Facebook, Instagram and Threads means, what triggers it, what Meta says about its limits, and what it does not prove.
• Meta's "AI info" labels on ads: what changed on June 1, 2026 and what an ad label does not tell youMeta's AI info labels on ads: what changed on June 1, 2026, which signals Meta says it reads, and what an ad label does not tell you about the picture.
• Platform labels compared: who labels AI images, how they decide, and what each label provesA table of how Meta, TikTok, YouTube and LinkedIn say they label AI images, what triggers each label, and what each one proves, from their own pages.
• YouTube's altered or synthetic content label: when it appears and what it meansWhen YouTube shows an altered or synthetic content label, who sets it, and the cases where YouTube says it may add one itself, including C2PA metadata.