Face Description

The face description node produces a natural-language description of the faces detected in an asset, using a vision-language model. It runs after Face Detection and skips work when no faces were found.

Kind

facedescription

Applies to

Image (video is stubbed / not yet implemented)

Inputs

Upstream facedetect output (reads face_count to skip when zero)

Output keys

face_description (String)

Requirements

A vision-language / LLM model service reachable from the worker. CPU on the worker; the model runs in the service.

Persists to

asset_json_comp + asset_node_result ledger

Configuration

Uses the model-service connection shared with the LLM/captioning family. The node only acts on assets where the upstream face detector reported at least one face.

Use Cases

  • Searchable people descriptions — "man with glasses, outdoors" style text for retrieval.

  • Accessibility / alt text for images containing people.

  • Review context — a quick textual summary of who/what is in a flagged image.