Project page: <a href=\"https://yihao-meng.github.io/CausalCine/\" rel=\"nofollow\">https://yihao-meng.github.io/CausalCine/</a></p>\n","updatedAt":"2026-05-13T02:05:28.661Z","author":{"_id":"656084f44e8918182d4f07c8","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/akAvCUCi7eR31PWOXrVPw.jpeg","fullname":"Yihao Meng","name":"Yhmeng1106","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":1,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.41350772976875305},"editors":["Yhmeng1106"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/akAvCUCi7eR31PWOXrVPw.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2605.12496","authors":[{"_id":"6a03dc0586b054ce2fa40d2e","name":"Yihao Meng","hidden":false},{"_id":"6a03dc0586b054ce2fa40d2f","name":"Zichen Liu","hidden":false},{"_id":"6a03dc0586b054ce2fa40d30","name":"Hao Ouyang","hidden":false},{"_id":"6a03dc0586b054ce2fa40d31","name":"Qiuyu Wang","hidden":false},{"_id":"6a03dc0586b054ce2fa40d32","name":"Ka Leong Cheng","hidden":false},{"_id":"6a03dc0586b054ce2fa40d33","user":{"_id":"662128ec9ca2cd4e6db2fb44","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/662128ec9ca2cd4e6db2fb44/uUg1V-pVfxT3mLuFgJuAN.jpeg","isPro":false,"fullname":"Bruce Yu","user":"bruceyyu","type":"user","name":"bruceyyu"},"name":"Yue Yu","status":"claimed_verified","statusLastChangedAt":"2026-05-13T07:51:06.090Z","hidden":false},{"_id":"6a03dc0586b054ce2fa40d34","name":"Hanlin Wang","hidden":false},{"_id":"6a03dc0586b054ce2fa40d35","name":"Haobo Li","hidden":false},{"_id":"6a03dc0586b054ce2fa40d36","name":"Jiapeng Zhu","hidden":false},{"_id":"6a03dc0586b054ce2fa40d37","name":"Yanhong Zeng","hidden":false},{"_id":"6a03dc0586b054ce2fa40d38","name":"Xing Zhu","hidden":false},{"_id":"6a03dc0586b054ce2fa40d39","name":"Yujun Shen","hidden":false},{"_id":"6a03dc0586b054ce2fa40d3a","name":"Qifeng Chen","hidden":false},{"_id":"6a03dc0586b054ce2fa40d3b","name":"Huamin Qu","hidden":false}],"mediaUrls":["https://cdn-uploads.huggingface.co/production/uploads/656084f44e8918182d4f07c8/nIhe0C3LtpuH1YhrCtN0T.mp4"],"publishedAt":"2026-05-12T00:00:00.000Z","submittedOnDailyAt":"2026-05-13T00:00:00.000Z","title":"CausalCine: Real-Time Autoregressive Generation for Multi-Shot Video Narratives","submittedOnDailyBy":{"_id":"656084f44e8918182d4f07c8","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/akAvCUCi7eR31PWOXrVPw.jpeg","isPro":false,"fullname":"Yihao Meng","user":"Yhmeng1106","type":"user","name":"Yhmeng1106"},"summary":"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/","upvotes":20,"discussionId":"6a03dc0686b054ce2fa40d3c","projectPage":"https://yihao-meng.github.io/CausalCine/","ai_summary":"CausalCine enables interactive, multi-shot video generation by addressing limitations of autoregressive models through causal modeling, dynamic memory routing, and real-time distillation techniques.","ai_keywords":["autoregressive video generation","causal modeling","multi-shot video generation","interactive generation","Content-Aware Memory Routing","causal base model","distillation","attention-based relevance scores","cross-shot coherence","real-time generation"],"organization":{"_id":"67c1d682826160b28f778510","name":"antgroup","fullname":"Ant Group","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/662e1f9da266499277937d33/7VcPHdLSGlged3ixK1dys.jpeg"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"656084f44e8918182d4f07c8","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/akAvCUCi7eR31PWOXrVPw.jpeg","isPro":false,"fullname":"Yihao Meng","user":"Yhmeng1106","type":"user"},{"_id":"662128ec9ca2cd4e6db2fb44","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/662128ec9ca2cd4e6db2fb44/uUg1V-pVfxT3mLuFgJuAN.jpeg","isPro":false,"fullname":"Bruce Yu","user":"bruceyyu","type":"user"},{"_id":"68a2fb20b3ad3d518527c49d","avatarUrl":"/avatars/cb37f5b2a98cbbea6f9726552a8f60e2.svg","isPro":false,"fullname":"zq","user":"capturee","type":"user"},{"_id":"620783f24e28382272337ba4","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/620783f24e28382272337ba4/zkUveQPNiDfYjgGhuFErj.jpeg","isPro":false,"fullname":"GuoLiangTang","user":"Tommy930","type":"user"},{"_id":"64d4a858a768729ccfd1ac06","avatarUrl":"/avatars/96152cd450a53a35d656f37c3ed419cc.svg","isPro":false,"fullname":"Tianrui_Feng","user":"jerryfeng","type":"user"},{"_id":"64acd2ec39fcfebff8c79c00","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/64acd2ec39fcfebff8c79c00/Avq66l5hO-aggNtk4Y1ss.png","isPro":false,"fullname":"Ka Leong Cheng","user":"felixcheng97","type":"user"},{"_id":"64981bea09cea550852652af","avatarUrl":"/avatars/df528e9008972c8e5ae4d278e617476c.svg","isPro":false,"fullname":"Qiuyu Wang","user":"qiuyuu","type":"user"},{"_id":"6478a982256b62e219917d67","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/PUJ-N2cQxgEmDGfyjajyA.jpeg","isPro":false,"fullname":"JingyeChen22","user":"JingyeChen22","type":"user"},{"_id":"68489221863b75bfb2097c8e","avatarUrl":"/avatars/e3d78fea4ca2e1ddb9e4bb03cb452929.svg","isPro":false,"fullname":"sigma","user":"sigma7863","type":"user"},{"_id":"655c2cc7e8a8971e89aa6e75","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/655c2cc7e8a8971e89aa6e75/rWQEbTqgq7MLEJJGfjP5y.jpeg","isPro":true,"fullname":"LIU Zichen","user":"LiuZichen","type":"user"},{"_id":"6617f604625ae15f6464113e","avatarUrl":"/avatars/e8f86a7c27bd0f0561fce4046d9e4c0a.svg","isPro":false,"fullname":"Yichong Lu","user":"Louischong","type":"user"},{"_id":"63f0baf66309c84d5f4a2226","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63f0baf66309c84d5f4a2226/ihOgtwseRkfP1t-60IgyT.jpeg","isPro":true,"fullname":"Qingyan","user":"QingyanBai","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"67c1d682826160b28f778510","name":"antgroup","fullname":"Ant Group","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/662e1f9da266499277937d33/7VcPHdLSGlged3ixK1dys.jpeg"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2605/2605.12496.md"}">
CausalCine: Real-Time Autoregressive Generation for Multi-Shot Video Narratives
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.
AI-generated summary
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/
Community
Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.
Tap or paste here to upload images
Cite arxiv.org/abs/2605.12496 in a model README.md to link it from this page.
Cite arxiv.org/abs/2605.12496 in a dataset README.md to link it from this page.
Cite arxiv.org/abs/2605.12496 in a Space README.md to link it from this page.
Discussion (0)
Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.
Sign in →No comments yet. Sign in and be the first to say something.