A training-free framework that reorders shot generation and maintains per-entity visual memory improves cross-shot character, object, and scene consistency over narrative-order memory baselines.
CausalCine: Real-Time Autoregressive Generation for Multi-Shot Video Narratives
2 Pith papers cite this work. Polarity classification is still indexing.
abstract
Autoregressive video generation aims at real-time, open-ended synthesis. Yet, cinematic storytelling is not merely the endless extension of a single scene; it requires progressing through evolving events, viewpoint shifts, and discrete shot boundaries. Existing autoregressive models often struggle in this setting. Trained primarily for short-horizon continuation, they treat long sequences as extended single shots, inevitably suffering from motion stagnation and semantic drift during long rollouts. To bridge this gap, we introduce CausalCine, an interactive autoregressive framework that transforms multi-shot video generation into an online directing process. CausalCine generates causally across shot changes, accepts dynamic prompts on the fly, and reuses context without regenerating previous shots. To achieve this, we first train a causal base model on native multi-shot sequences to learn complex shot transitions prior to acceleration. We then propose Content-Aware Memory Routing (CAMR), which dynamically retrieves historical KV entries according to attention-based relevance scores rather than temporal proximity, preserving cross-shot coherence under bounded active memory. Finally, we distill the causal base model into a few-step generator for real-time interactive generation. Extensive experiments demonstrate that CausalCine significantly outperforms autoregressive baselines and approaches the capability of bidirectional models while unlocking the streaming interactivity of causal generation. Demo available at https://yihao-meng.github.io/CausalCine/
fields
cs.CV 2years
2026 2verdicts
CONDITIONAL 2representative citing papers
An open-source causal video world model sustains hour-long, 720p/60fps interactive generation without visual drift, paired with a VLM-based director-pilot agentic harness for rich, open-ended interaction.
citing papers explorer
-
GroundShot: Visually Consistent Multi-Shot Long Video Generation via Entity-Grounded Shot Scheduling
A training-free framework that reorders shot generation and maintains per-entity visual memory improves cross-shot character, object, and scene consistency over narrative-order memory baselines.
-
Infinite Worlds with Versatile Interactions
An open-source causal video world model sustains hour-long, 720p/60fps interactive generation without visual drift, paired with a VLM-based director-pilot agentic harness for rich, open-ended interaction.