AI content and how we label it

Last updated: 30 October 2026

Every image, video and audio file made with Turbogen is created or edited by artificial intelligence. Following the EU AI Act (Regulation (EU) 2024/1689, art. 50) we make this visible to you and machine-readable for everyone who receives the file.

Visible labels

• Results in the studio carry an "AI-generated" badge and a note in the details view.
• Files sent by our Telegram bot say "AI-generated" in the caption.
• Examples on our public pages are marked as AI-generated.

Labels inside the file

When a file is delivered we add metadata without re-encoding the picture or sound:

• Images (PNG, JPEG, WebP): XMP with the IPTC field DigitalSourceType = trainedAlgorithmicMedia (generated), compositeWithTrainedAlgorithmicMedia (generated from your reference image) or algorithmicallyEnhanced (our upscale / restore tools), a description "AI-generated by Turbogen" and, in PNG, a text chunk with the same words.
• Video (MP4 / MOV): an XMP box with the same fields.
• Audio: MP3 ID3 tags (TXXX:AI-generated, TXXX:DigitalSourceType, comment) or WAV INFO / XMP chunks.

You can check a file with any metadata viewer, for example exiftool -XMP:all -PNG:all file.png.

Provenance from the model makers

Some models add their own provenance: Google models embed an invisible SynthID watermark, and some providers attach C2PA "Content Credentials". We never remove them. A file that already carries C2PA Content Credentials is delivered unchanged, because changing its bytes would break the signed credentials, which already declare it AI-generated. The "remove watermark" option and our object remover only touch visible pixels; invisible watermarks remain. We plan to add our own signed C2PA credentials as well.

What you need to do

• Do not remove or alter these labels.
• If you publish realistic AI images, video or audio of real people, places or events (deepfakes), clearly disclose that they are AI-generated or manipulated. For evidently artistic, satirical or fictional works a disclosure that does not spoil the work is enough.
• Follow our Acceptable Use rules: no sexual content, no "undress" edits, no sexual content involving minors or real people, no deceptive deepfakes.

Moderation

Prompts, lyrics and uploaded images are checked automatically before a job starts, in English, German, Spanish, Polish, French, Portuguese, Italian and Russian. Refused requests are not charged. Report content that breaks the rules or a missing label to [email protected].