detect ai image

This block is part of the files group and was last modified in core v15.0.1.

detect ai image condition v15.0.1

Reads what an image says about how it was made, without any external service: C2PA content credentials, the IPTC digital source type and AI-system fields, and the generation records Stable Diffusion tools write. Names the generator (for example GPT-4o, Adobe Firefly, Imagen 4 or a Stable Diffusion checkpoint) and its maker. Reads PNG, JPEG and WebP. A screenshot or re-encode strips this metadata, so ai generated off means nothing was found, not that a person made the image.

In pins 1 + flow

flow
flow
file required
file
The image to check. The first file is read.

Out pins 11 + flow

flow
flow
success
condition
The image was read. Says nothing about whether it is AI-made.
ai generated
condition
The metadata declares the image AI-generated or AI-edited, or carries a generator's own record. Off means nothing was found.
generator
value
The tool or model that made the image, as the metadata names it (GPT-4o, Adobe Firefly, Imagen 4, sd_xl_base_1.0, ComfyUI). Empty unless ai generated.
made by
value
The company behind the generator (OpenAI, Google, Adobe, Midjourney, Stability AI, …). Empty unless ai generated and the maker is known.
source type
value
The IPTC digital source type: trainedAlgorithmicMedia (generated), compositeWithTrainedAlgorithmicMedia (AI-edited), or a non-AI code such as digitalCapture. Empty when none is declared.
content credentials
condition
The image carries a readable C2PA manifest.
c2pa signed by (unverified)
value
The organisation in the active C2PA manifest's signing certificate. The signature is not validated, so a forged file can name anyone.
c2pa claim generator
value
The application the active C2PA manifest says wrote it.
evidence
array<object>
Every signal found, as {source, field, value}; source is c2pa, xmp, exif or png_text.
warnings
array<value>
Metadata that was present but unreadable, a stripped C2PA manifest, or a remote manifest that was not fetched.
error
value
Why the image could not be read when success is off.

Offline, metadata-only AI detection for an image file (PNG/JPEG/WebP). Makes NO network call and does not read SynthID watermarks (that needs Google's or OpenAI's servers). ai_generated=true when C2PA or XMP declares IPTC trainedAlgorithmicMedia / compositeWithTrainedAlgorithmicMedia, XMP names an AI system, a Stable Diffusion/ComfyUI/InvokeAI/NovelAI record is present, or a software tag names an AI generator. ai_generated=false means NO METADATA FOUND, never 'human-made'. source_type tells generated (trainedAlgorithmicMedia) from AI-edited (compositeWithTrainedAlgorithmicMedia) and also reports non-AI codes such as digitalCapture. C2PA signatures are NOT validated: c2pa_signed_by is the name in the certificate as written. success=false only when the file could not be read (MISSING_FILE, INVALID_FILE, UNSUPPORTED_FILE_TYPE, UNREADABLE_IMAGE); damaged metadata lands on warnings with success=true.

ai

ai image

c2pa

content credentials

provenance

generated

get takes the image from the form and detect AI image reads what its metadata says about how it was made, with no external service. When ai generated is on, the branch runs update document, which stores the generator and its maker through set fields. Off means no metadata was found, not that a person made the image: a screenshot or a re-encode strips it. The success pin says only that the file could be read.

Version history

Introduced in v15.0.1.

Back to files Return to the main group to view all sub-groups Back to images Return to the group to view all blocks within this group