The EU icon set distinguishes two disclosures: AI GENERATED for content that is fully AI-created, and AI MODIFIED for content that builds on existing material. Samsa decides between the two automatically, based on how each asset was actually made.
This is deliberate: the icon type is not a setting. Neither admins nor members pick it — so the label on a file always tells the truth about its origin, and it always matches what the file's Content Credential says. The two can never disagree.
An asset counts as fully AI-generated when it was created from your prompt alone — no source image and no reference image or video were involved. This includes generations that use:
Your trained models — Styles, Objects, Persons, and Scenes. A trained model shapes how the AI generates; it isn't source material being edited.
Your color palettes — same reasoning: a palette guides the generation, it isn't an input image.
So a typical on-brand workflow — a prompt plus your trained brand Style — produces content labeled AI GENERATED.
An asset counts as AI-modified when existing media went into it or it was derived from another asset. That covers:
Generations that used a reference image
Image-to-image creations
Edits: inpainting, outpainting, and variations
Upscaling — an upscaled version is a derivative, so it's labeled AI MODIFIED even if the image it's based on was fully generated
Background removal
Image-to-video — the start (or end) frame counts as source material
Videos created from existing footage or images
The official EU set also includes a basic "AI" icon. Samsa never selects it automatically — it's only used when explicitly requested through the API.
On the visible AI label when your organization uses labeled downloads
In the file's Content Credential, readable by verification tools such as detect.samsa.ai
In the Compliance tab (avatar menu > Organization > Compliance), where the EU icons are shown together with an explanation of when each one is used
