A-Harness

An integrated visual detection and segmentation tool that combines open-vocabulary object detection with prompt-driven precise segmentation capabilities, implementing a complete recognition pipeline from textual descriptions to pixel-level masks.

An imagination-driven prompt writer that visualizes plausible physical interactions within an image to enhance contextual and functional reasoning for editing tasks.

Web search tool that retrieves concise textual knowledge for an affordance task: object parts, affordance mechanism, and how humans interact with the object. The decision model should CUSTOMIZE the search by providing a search strategy (search_focus, target_part, interaction_type) based on its analysis of the task. This produces much more targeted and useful results than a generic search. Returns: affordance_name, part_name, object_name, reasoning (text insights), plus searchable evidence (queries, results, crawled URLs).

refine ambiguous recognition by zooming into a selected region for closer analysis