C2PA · Stable Diffusion · Generator tags
AI image metadata remover
A free AI metadata remover for generated images. See the prompts, seeds, model names and signed provenance records yours carries — then decide what to do about them.
Where AI image generators hide their metadata
There is no single standard, so the data ends up in several different places:
- PNG text chunks. Automatic1111, ComfyUI and Forge write a
parameterschunk containing the complete prompt, negative prompt, seed, sampler, CFG scale, model name and model hash. Anyone who opens the file in a text editor can read it. - C2PA manifests. A JUMBF box holding a cryptographically signed provenance record. In JPEG it lives in an
APP11segment, in PNG acaBXchunk, in HEIC and AVIF ajumbbox. - EXIF and XMP fields. Generator names appear in
Software,UserCommentorCreatorTool, plus custom XMP namespaces.
Your prompt is often the most sensitive part. It can contain client names, unreleased product details, or simply a working method you would rather not publish.
What C2PA is, and what removing it means
C2PA Content Credentials are a signed record of how a file was made — which tool, when, and whether AI was involved. They are not a watermark and not DRM. They are closer to a chain of custody, and the signature means tampering is detectable rather than prevented.
Removing a manifest removes the signed history. A verifier then reports the file as having no credentials, which is the same state as any ordinary photo. It does not make the file appear to be something it is not.
There are honest reasons to strip them. A credential can name the account, the tool and the device you used. If you have licensed an image to a client, the embedded chain can also expose your workflow. Read more in C2PA explained, or go straight to the ChatGPT image metadata remover if that is where your images come from.
Posting to Instagram, Facebook or Threads? What triggers Meta's AI info label explains which of its signals live in the file and which do not, and SynthID vs C2PA covers the difference between a label in the file and a watermark in the pixels.
Where we draw the line
This tool removes metadata. It does not modify your pixels, and it will not implement fingerprint-scrambling designed to defeat AI classifiers. That feature exists elsewhere; we think it makes your image worse in exchange for helping mislead a system built to label content accurately.
Working through a folder of generations rather than one file? The bulk image metadata remover handles them in a single pass.
Be aware of the surrounding rules too. The EU AI Act's Article 50 transparency obligations apply from 2 August 2026, and most large platforms already ask for AI content to be disclosed. Cleaning metadata for privacy is a normal thing to do. Removing a disclosure to pass work off as something it is not is a different act with different consequences.
Questions
Frequently asked questions
What metadata do AI image generators leave behind?
Does removing this make my image undetectable as AI?
Is it legal to remove AI metadata from my own images?
Do you modify pixels to defeat detection?
Questions people ask about this
- Does removing metadata make an AI image undetectable?No. Metadata is a label attached to the file. Detection largely works on the pixels themselves and on watermarks embedded in them, and stripping metadata touches neither.
- What metadata do ChatGPT images contain?Principally C2PA content credentials — a signed record marking the image as AI-generated — alongside ordinary format-level fields such as dimensions and colour profile.