{"slug": "mvagent-multi-agent-video-generation-via-consistent-condition-construction-and", "title": "MVAgent: Multi-Agent Video Generation via Consistent Condition Construction and Shot-Level Policy Optimization", "summary": "Researchers introduced MVAgent, a multi-agent pipeline that recasts multi-shot agentic video generation as condition construction, with agents collaborating through typed conditioning inputs. MVAgent uses a Spatial Grounding agent to anchor each shot to a matching camera-traversal view, an Observer that records shot endings in a continuity memory, a Transition agent that builds character action and spatial references for the next shot, and an Orchestrator that composes these inputs into each request, trained via agentic reinforcement learning with Trunk-GDPO, which compares rendered candidates at every shot rather than once per video. With generator and judges frozen, MVAgent attained the highest cross-shot consistency and narrative-planning quality among compared methods on ViMax-Bench and was preferred over the strongest agentic baseline in human evaluation.", "body_md": "arXiv:2609.30609v1 Announce Type: new \nAbstract: Multi-shot agentic video generation requires consistent character appearance, stable spatial layout across camera angles, and continuous character state between shots. When every shot is a separate request to a frozen generator, repeated text does not determine appearance, layout or state. We therefore recast the problem as condition construction and present MVAgent, a multi-agent pipeline whose agents collaborate through typed conditioning inputs. Because an environment image shows one viewpoint, a Spatial Grounding agent samples views from generated camera-traversal clips and anchors each shot to the view matching its framing. As generated shots drift from the plan, an Observer records how each shot ends in a continuity memory, from which a Transition agent builds character action and spatial references for the next shot. An Orchestrator composes these inputs into each request. Since a request reveals its effect only after rendering, we train it by agentic reinforcement learning with Trunk-GDPO, which compares rendered candidates at every shot rather than once per video and continues the best as the trunk. With generator and judges frozen, MVAgent attains the highest cross-shot consistency and narrative-planning quality among the compared methods on ViMax-Bench and is preferred over the strongest agentic baseline in human evaluation.", "url": "https://wpnews.pro/news/mvagent-multi-agent-video-generation-via-consistent-condition-construction-and", "canonical_source": "https://arxiv.org/abs/2609.30609", "published_at": "2026-09-28 04:00:00+00:00", "updated_at": "2026-09-28 04:20:39.958620+00:00", "lang": "en", "topics": ["artificial-intelligence", "generative-ai", "ai-agents", "ai-research", "computer-vision"], "entities": ["MVAgent", "Trunk-GDPO", "ViMax-Bench", "Spatial Grounding agent", "Observer", "Transition agent", "Orchestrator"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/mvagent-multi-agent-video-generation-via-consistent-condition-construction-and", "markdown": "https://wpnews.pro/news/mvagent-multi-agent-video-generation-via-consistent-condition-construction-and.md", "text": "https://wpnews.pro/news/mvagent-multi-agent-video-generation-via-consistent-condition-construction-and.txt", "jsonld": "https://wpnews.pro/news/mvagent-multi-agent-video-generation-via-consistent-condition-construction-and.jsonld"}}