Text to Image by Person

Executes a combined semantic and person-filtered search that takes a natural language text query (e.g., 'Jane Doe at a press conference' or 'John Smith speaking at a podium') and searches within a dataset to find images or video keyframes that both match the text query AND contain a specific enrolled person. The person must be enrolled in the organization for detection in images and video frames. You must specify the person via either person_ids (unique identifiers) OR person_names_or_aliases (names/aliases of enrolled persons) - exactly one is required, not both. The text query is encoded using the dataset's vision-language encoder (CLIP, SigLIP, Perception Encoder, etc.) and searched via vector similarity, filtered to assets where the specified person appears. Works with both legacy and native video model datasets. Use asset_type='image' for still images or 'keyframe' for video frames. Returns a list of matching image/keyframe results (each with asset ID, dataset ID, asset type, and for keyframes: video ID, composite slice ID and composite type) plus a list of video IDs aggregated from matching keyframes. Optional moderation scores are included per asset if requested.