Controlled AI Video, Human-Led Production, and Agentic Post

Seedance 2.5 and adjacent workflows are moving AI video toward more controllable cinematography through reference separation, character sheets, multimodal assets, continuity tools, 3D previs, and node-based orchestration. Feature-length Chinese and hybrid productions, Resolve’s MCP server, and integrated virtual-production cases suggest movement beyond novelty, but compute costs, usable-shot rates, continuity, human performance, permissions, provenance, and independent economic benchmarking remain unresolved; recent tool and partnership pieces remain promotional rather than decisive evidence.