NewThe detectors that scored perfect collapsed the hardest under attack.

Benchmarks

The data is organized into benchmarks. A benchmark is a versioned dataset (for example synthetic-face-v1) with its own set of queryable dimensions, the same way different models expose different parameters. You choose a benchmark, discover its dimensions, then query items within it. Every request that touches data takes a benchmark parameter.

List the benchmarks your key can query with /api/v1/data/benchmarks, then pass ?benchmark=<id> to /catalog (to see its dimensions and values) and /items (to select images). The dimensions differ per benchmark; /catalog is always the source of truth for what a given benchmark supports.

NOTE
Test tier (free sample). A test key pulls a fixed free sample instead of the full corpus: one real and one fake per demographic cell. For passport-pad-v1 that is 12 identity-matched real and fake pairs across the cells, from both editors; synthetic-face-v1 adds a set of platform-condition examples on top. Live keys see the full corpus, metered by credits.
NOTE
Choosing a benchmark. If your key sees exactly one benchmark you may omit the benchmark parameter and it is used by default. Once your key can see more than one, the parameter is required and omitting it returns a 400 listing the available ids. New benchmarks are added without changing this contract, so integrations built against one benchmark keep working.

Available benchmarks

Every benchmark carries a stable code (for example MG-SYN-01) alongside its versioned id (synthetic-face-v1). The benchmark parameter accepts either. The id stays the canonical value echoed back in responses; the code is the citable alias. Expand a tab to see what a benchmark contains and the dimensions it exposes. For full dataset specs (composition, image spec, labeling) open its data card.

MG-SYN-01Availablesynthetic-face-v1test, live

Real (genuine, unmodified) portraits paired with identity-conditioned synthetic faces across demographic cells and platform conditions. Real and generated items are linked by lineage: a sourced real image plus everything derived from it, all sharing one source_real_id.

Contents
clean images27,997
conditions18
demographic cells12
pairinglineage by source_real_id
generatorsSDXL with InstantID
Query dimensions
kindskin_tonegendergeneratorperturbationlayerscenebase_idsource_real_idttp_tacticttp_technique
Catalog endpoint
GET /api/v1/data/catalog?benchmark=synthetic-face-v1
Data cardEndpoints

Live counts and dimensions are returned by GET /api/v1/data/benchmarks and /api/v1/data/catalog.

PreviousRate limitsNextSelecting images