metadata
title: AV Curation Pipeline — e2e run (9NcjRFu6C4I)
emoji: 🚂
colorFrom: blue
colorTo: green
sdk: static
pinned: false
AV curation agentic pipeline — end-to-end run on 9NcjRFu6C4I
Step-by-step input/output visualization of one full LangGraph pipeline run: Qwen3-Omni caption → sounding-object → SAM3 segmentation → EffectErase inpainting → visual check → SAM-Audio best-of-10 (visual+text prompts × 5 seeds, ImageBind-ranked) → paired AV output. Includes all 10 audio candidates with their scores and the demo gate overrides used for this hard (synthetic) clip.