AI transparency

• Marker in the file — generated images, video and audio carry a machine-readable record in their metadata stating that they were created with AI (IPTC standard trainedAlgorithmicMedia: EXIF and XMP in images, XMP in video, ID3 or file metadata in audio).
• File name — every downloaded file has the ai-generated tag in its name.
• Provider markings stay — if a model attaches its own content credentials (C2PA) or an invisible watermark (e.g. SynthID in Google models), we do not remove them.
• Nothing on the content itself — we do not burn watermarks or captions into images and videos.
• In preparation — C2PA content credentials and an invisible watermark for models that do not have one, so that the marking also survives the file being saved again.

• Prompt filters in many languages — we also check the text after translating it into English.
• Automatic analysis of uploaded files and results for nudity; in videos, we analyse the frames. The model providers’ safety filters are also enabled.
• Features that process a likeness or voice (e.g. face swap, voice clone) require a declaration that you have the right to that likeness or voice.
• Publications in the Gallery are checked by a human. Anyone — including people without an account — can report content with the “Report” button; we process reports of non-consensual intimate images and of content concerning minors within 48 hours.
• We block accounts that break the rules and report serious cases to the competent authorities.

If you publish or share content that realistically depicts existing people, places or events and could be taken as genuine (e.g. face swap, talking avatar, voice clone), you act as a deployer within the meaning of the AI Act and must disclose that the content was generated or manipulated by AI. For evidently artistic, satirical or fictional works, a disclosure that does not hamper the enjoyment of the work is sufficient.

In ads, do not present an AI avatar as a real customer or its words as a genuine review. Label the ad as generated using AI where the law or the platform’s rules require it (e.g. TikTok and Meta). Details in §11 of the Terms.

• Non-consensual intimate imagery (NCII) of real people, including “undressing” people in photos and videos.
• Any sexual content involving minors (CSAM) — zero tolerance.
• Deepfakes of real people created without their consent to mislead, defame or defraud.

These uses are prohibited by our Terms and may constitute a criminal offence.

Fattly integrates AI models from third-party providers and does not guarantee a specific result. AI-generated content may be similar to other users’ results. Your rights and obligations are described in the Terms and the Privacy Policy.

Questions about AI transparency or the provenance of content: contact@fattly.app.