Transparency
AI Disclosure
How CladeGrove marks AI-generated images, what every output file contains, and what this means for anyone who receives one.
Last updated: April 27, 2026
1. What CladeGrove makes
CladeGrove is an AI photo studio. Every image produced by the Service is a synthetic image generated by a third-party AI model on the basis of a textual persona specification. The intent of the product is to depict fictitious personas: not photographs of real human beings, not face-clones of real human beings, and not biometric replicas of real human beings.
As of 27 April 2026, the Service no longer accepts face photos of identifiable real persons as character source material; new characters can only be created from text (modes text_free and text_guided).
Caveat: legacy characters. Characters created before 27 April 2026 may have an associated stored face reference photo. When a user tags such a legacy character in a new generation, that stored photo is sent to the AI image model as an identity reference. The output is still produced by the AI model (it is not a pixel-faithful copy), but the identity is anchored to a real photograph rather than fully fabricated from a textual description. The user can delete the legacy character (and the stored photo) at any time. See Privacy Policy §2 for the full treatment.
See our Terms of Service §5 for the Acceptable Use rules that prohibit attempts to clone, impersonate or generate non-consensual imagery of real persons.
2. How outputs are marked (Art. 50 EU AI Act)
Pursuant to Article 50(2) of Regulation (EU) 2024/1689 (the EU AI Act), generated outputs must be "marked in a machine-readable format and detectable as artificially generated or manipulated". Before persistence, every Synthetic Output produced by CladeGrove is stamped with two layers of provenance metadata:
2.1 EXIF (IFD0)
Software:CladeGrove · https://cladegrove.comArtist:AI-generated by CladeGroveCopyright:AI-generated content. Synthetic persona. © {year} CladeGroveImageDescription:AI-generated synthetic image. Not a real person.
2.2 XMP packet (IPTC + custom GenAI namespace)
The XMP packet is the legally relevant marker. It carries the IPTC controlled-vocabulary value endorsed by the European Commission AI Office and IPTC for synthetic-media labelling:
Iptc4xmpExt:DigitalSourceType =
http://cv.iptc.org/newscodes/digitalsourcetype/trainedAlgorithmicMediaplus a custom GenAI: namespace recording:
GenAI:Provider: model and platform (literal valueOpenAI gpt-image-2 via CladeGrove);GenAI:GenerationId: internal stable identifier of the generation;GenAI:CreatedAt: ISO-8601 timestamp of generation;GenAI:TermsURL: URL of the Terms of Service;GenAI:DisclosureURL: URL of this AI Disclosure page;GenAI:Disclaimer: literal value "Synthetic image. Does not depict a real person."
The XMP packet also carries standard fields confirming AI origin: xmp:CreatorTool = CladeGrove, dc:rights = AI-generated content, xmpMM:DerivedFrom = synthetic-prompt, photoshop:Credit = Generated with AI · CladeGrove.
2.3 Roadmap: C2PA Content Credentials
We are committed to upgrading provenance to a cryptographically signed C2PA Content Credentials manifest before the date of applicability of Art. 50 (2 August 2026). C2PA produces a tamper-evident manifest that survives re-encoding by most professional editing tools and can be verified by third-party inspectors (e.g. Content Credentials browser extensions, newsroom verification pipelines).
3. How to inspect the markers
You can verify the AI-generated provenance of any file produced by CladeGrove using standard tools:
- macOS: open the image in Preview → Tools → Show Inspector →Exif / IPTC tab.
- Windows: right-click the image → Properties → Details.
- Cross-platform: ExifTool (run
exiftool image.pngto print every EXIF and XMP field). - Online: any public XMP/EXIF inspector (no upload of sensitive personal data is required, since CladeGrove outputs do not depict real persons).
4. Caveat: third-party platforms strip metadata
Several social platforms, messaging apps, image-compression CDNs and screenshot tools strip EXIF and XMP metadata when re-encoding images. Where this happens, the machine-readable markers described above may be lost. Loss of metadata does not change the AI-generated nature of the content; it changes only the ability of automated tools to detect it.
This is one of the reasons we plan to add C2PA: C2PA manifests are cryptographically signed and survive re-encoding by C2PA-aware tools. Until that ships, you should not rely on EXIF/XMP alone if the image has passed through a third-party platform.
5. Obligations of users who publish outputs (Art. 50(4) AI Act)
When a user of CladeGrove publishes, shares or otherwise distributes a Synthetic Output to third parties, the user becomes a deployer under Art. 50(4) of the AI Act. Where the output could be perceived as authentic photography of a real person, situation or event, the deployer is responsible for labelling the content as artificially generated or manipulated in a clear and distinguishable manner, at the latest at the time of the first interaction or exposure.
The provenance metadata we embed assists, but does not replace, that user-side disclosure obligation. In contexts where metadata is likely to be stripped (most social platforms, messaging), users should add a visible disclosure: a caption ("AI generated"), an on-image label or a watermark.
See Terms of Service §7 for the full deployer clause.
6. Reporting and right to object
If you believe a Synthetic Output produced by CladeGrove incidentally resembles you to a degree that is recognisable to a third party, or if you wish to report a suspected violation of our Acceptable Use rules (impersonation, NCII, deepfake, content depicting minors, identity replication of a real person), write to privacy@cladegrove.com or to support@cladegrove.com. Provide a description of the relevant generation (and, where available, the URL or file). We will remove the content within 7 days of verification and, in cases of policy violation, terminate the offending account.
7. Updates
This page is updated whenever the technical marking of outputs changes. The current implementation is described in src/lib/image-metadata.ts in our codebase and reflected verbatim above.