CausalCine: Real-Time Autoregressive Generation for Multi-Shot Video Narratives

Hugging Face Daily Papers Papers
video-generation autoregressive-models real-time-generation multi-shot-narratives causal-modeling

Summary

CausalCine is a new academic framework for real-time, interactive multi-shot video generation that uses causal modeling and dynamic memory routing to improve cross-shot coherence in autoregressive models.

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/
Original Article
View Cached Full Text

Cached at: 05/13/26, 04:12 AM

Paper page - CausalCine: Real-Time Autoregressive Generation for Multi-Shot Video Narratives

Source: https://huggingface.co/papers/2605.12496 Authors:

,

,

,

,

,

,

,

,

,

,

,

,

Abstract

CausalCine enables interactive, multi-shot video generation by addressing limitations of autoregressive models through causal modeling, dynamic memory routing, and real-time distillation techniques.

Autoregressive video generationaims 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 transformsmulti-shot video generationinto 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 acausal base modelon native multi-shot sequences to learn complex shot transitions prior to acceleration. We then proposeContent-Aware Memory Routing(CAMR), which dynamically retrieves historical KV entries according toattention-based relevance scoresrather than temporal proximity, preservingcross-shot coherenceunder bounded active memory. Finally, we distill thecausal base modelinto a few-step generator for real-timeinteractive 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/

View arXiv pageView PDFProject pageAdd to collection

Get this paper in your agent:

hf papers read 2605\.12496

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2605.12496 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2605.12496 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2605.12496 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Long Video Generation (4 minute read)

TLDR AI

The article introduces A²RD, a novel architecture for generating consistent long videos using agentic autoregressive diffusion. It proposes a Retrieve–Synthesize–Refine–Update cycle and a new benchmark, LVBench-C, to address semantic drift in long-horizon video synthesis.