Soap2Soap: Long Cinematic Video Remaking via Multi-Agent Collaboration

We study series-level cinematic remaking, a long-horizon video-to-video generation problem that localizes full episodes or films via stylization or actor replacement while strictly preserving narrative structure, motion choreography, and character identity across hundreds of shots. Existing video generation and editing pipelines often break down in this regime due to compounding identity drift, background mutation, and semantic erosion under large camera motions and viewpoint changes. We propose Soap2Soap, a multi-agent framework that enforces long-term language-visual consistency through a Dual-Bridge Consistency mechanism: a scene-aware JSON screenplay serving as a persistent semantic backbone, and dynamically allocated visual reference anchors at both scene and shot levels. To suppress drift before video synthesis, we introduce batch keyframe consistency, jointly generating multiple keyframes in a shared latent context via a grid-based formulation. A closed-loop verification agent further audits identity, stability, and alignment to trigger selective regeneration. Experiments on SoapBench demonstrate strong improvements over commercial video generation APIs in long-term consistency and narrative fidelity.

📋 한 줄 요약

**[Long-Form Video Generation / Multi-Agent]** Soap2Soap이 JSON screenplay·dynamic visual anchor·batch keyframe·closed-loop verification으로 episode 단위 long-horizon cinematic remaking 가능, SoapBench에서 commercial API 능가.

🎯 핵심 기여도

Series-level cinematic remaking — 전체 episode·film을 stylization·actor replacement로 localize하면서 narrative structure·motion choreography·character identity를 hundreds of shot에 걸쳐 strictly 보존하는 long-horizon video-to-video 생성 문제 정의.
기존 video generation·editing 파이프라인이 이 regime에서 compounding identity drift·background mutation·semantic erosion으로 break down(큰 카메라 동작·viewpoint 변화 하) 진단.
Soap2Soap 제안 — multi-agent 프레임워크로 Dual-Bridge Consistency 메커니즘을 통해 long-term language-visual consistency 강제: scene-aware JSON screenplay(persistent semantic backbone)와 scene·shot 두 수준에서 dynamically 할당된 visual reference anchor.
Video 합성 전 drift 억제 위해 batch keyframe consistency 도입(grid-based formulation으로 multiple keyframe을 shared latent context에서 jointly 생성), closed-loop verification agent가 identity·stability·alignment 감사해 selective regeneration trigger.

💡 핵심 아이디어

Episode 단위 long-horizon cinematic remaking의 핵심 난제는 hundreds of shot에 걸친 identity·narrative drift이며, persistent semantic backbone(JSON screenplay) + dynamic visual anchor의 dual-bridge + 합성 전 batch keyframe 일관성 + closed-loop verification의 multi-agent 협업으로 해결 가능하다.

🔬 기술적 접근법

**방법론**: Soap2Soap — multi-agent framework with Dual-Bridge Consistency.
**핵심 기법**: (1) Scene-aware JSON screenplay를 persistent semantic backbone으로, (2) Scene·shot 두 수준에서 dynamically 할당된 visual reference anchor, (3) Batch keyframe consistency — grid-based formulation으로 multiple keyframe을 shared latent context에서 jointly 생성, (4) Closed-loop verification agent가 identity·stability·alignment 감사, (5) Selective regeneration trigger로 drift 발생 시 대응.

📊 주요 결과

SoapBench에서 commercial video generation API 대비 long-term consistency·narrative fidelity 강한 향상.
Hundreds of shot에 걸친 character identity·motion choreography 보존 입증.
큰 카메라 동작·viewpoint 변화 하에서도 background mutation·semantic erosion 억제.
Multi-agent 협업이 단일 모델 대비 long-horizon 안정성에서 우월.

💭 의의 및 한계

**의의**: Series-level cinematic remaking이라는 새 long-horizon 과제 정의, multi-agent 협업의 video generation 응용 모범 사례, JSON screenplay·dual-bridge·closed-loop verification의 조합 레시피 정립. **한계**: 합성 비용·복잡도 증가, SoapBench 단일 벤치마크 평가, JSON screenplay 자동 추출 품질 의존, 매우 미세한 actor expression·lip-sync 등은 추가 검증 필요.

🚀 실용적 활용

영화·드라마 시리즈 자동 remaking·dubbing·번역.
스타일·actor 변경 cinematic editing.
Long-form video generation의 일반 consistency framework.