Signed OpenAI claim
A readable manifest can identify the claim generator and declared provenance associated with the asset.
DALL·E 3 / OpenAI provenance
Inspect OpenAI image provenance locally
Read C2PA separately from ordinary text, EXIF and XMP. Create a derivative without supported embedded sections while keeping the original intact.
CONTEXT / EVIDENCE
OpenAI documents provenance, not a guaranteed prompt field
OpenAI says images generated with ChatGPT, Codex and its API include C2PA metadata. The signed claim is a provenance signal and can be removed accidentally or intentionally, so its absence does not establish how an image was made.
C2PA does not mean that a complete private prompt or seed is guaranteed to be stored as readable text. ClearTags reports what this particular file contains and can remove supported embedded metadata from a new publishing copy.
Keep three signals separate
C2PA · EXIF · XMP · CONTAINERPrompt-bearing fields
EXIF, XMP, PNG text and comments are inspected separately because a full prompt is not guaranteed by C2PA.
Non-metadata signals
Metadata cleanup does not remove a signal embedded in image content itself, including an invisible watermark.
Create a metadata-clean derivative
WORKFLOW / LOCAL COPY- 01
Keep the downloaded original
Preserve the source image if provenance or generation history may matter later.
- 02
Inspect the exact export
Review C2PA, EXIF, XMP and text fields in the asset you plan to publish.
- 03
Select embedded sections
Remove the supported provenance or plain-text metadata you do not intend to distribute.
- 04
Reinspect the new copy
Confirm the report after cleanup rather than assuming every provenance mechanism was removed.