Images from ChatGPT, Midjourney, DALL·E, and Adobe Firefly carry invisible C2PA credentials and XMP tags that name the model — sometimes the exact prompt — used to create them. Blindfold finds and removes these markers entirely in your browser. No upload, ever.
Removing metadata is not the same as defeating every AI detector. Blindfold strips the metadata-based markers below, which is the layer most tools, platforms, and casual inspection rely on. It cannot remove pixel-level watermarks like Google's SynthID, which some providers now bake directly into the image data — those survive metadata removal because they were never metadata to begin with.
The Coalition for Content Provenance and Authenticity standard that ChatGPT, Adobe Firefly, and other major generators now embed by default. It records which model made the image, when, and cryptographically signs the claim — acting like a nutrition label that anyone can read.
Some tools write the generation prompt directly into the file's metadata. If you generated something you'd rather not have attributed to you word-for-word, that text is sitting in the file unless it's stripped.
A broader Adobe-originated metadata format that can carry authorship, editing history, and — for AI tools — generation details, independently of the C2PA manifest.
Simple identifying tags naming the specific tool or model version used, often left behind even by generators that don't implement full C2PA.
Marketers and content creators uploading AI-assisted work to stock platforms or client deliverables, where undisclosed generation metadata can trigger a rejection or a policy violation.
Anyone who used a prompt they'd rather not have attached to the file — some generators write it straight into the metadata.
Freelancers delivering AI-generated assets who want to hand over a clean file without a visible paper trail of exactly how it was made.
No, and any tool that claims this isn't being straight with you. C2PA and XMP metadata are the easiest signal to strip — Blindfold removes them for Pro users. But some providers, including Google and OpenAI, now also embed invisible watermarks such as SynthID directly into the image pixels. Those survive metadata removal, re-saving, and screenshots, because they aren't metadata at all — they're part of the image data itself. Blindfold cleans metadata, not pixels.
C2PA Content Credentials (which record the generating model, and sometimes the exact prompt), XMP provenance tags, and generator signatures embedded by tools like ChatGPT, DALL·E, Midjourney, and Adobe Firefly. Free accounts can see these markers on any image; removing them is a Pro feature.
Some generators write the exact prompt into the image's metadata as part of its provenance record — intended as a transparency feature, but it means anyone who downloads the file can read your prompt in a metadata viewer, even if you never intended to share it.
Increasingly, yes. Stock platforms scan for C2PA credentials to enforce AI-content policies, and some clients now check metadata before accepting deliverables. An image that visually looks fine can still be flagged the moment someone inspects its metadata.
Detecting and viewing AI markers is free. Removing them is part of Blindfold Pro ($5/month), alongside PDF/DOCX/PPTX/XLSX support and batch processing.