Hugging Face Daily Papers · August 21, 2026 · 4 min read

4DAnyone: Create Anyone in 4D from a Casual Monocular Video

Mirrored from Hugging Face Daily Papers for archival readability. Support the source by reading on the original site.

Like Read original ↗

4DAnyone turns a casual monocular video into multi-view videos, enabling downstream 4DGS reconstruction.\n","updatedAt":"2026-08-21T02:30:15.222Z","author":{"_id":"649958942ca6f96c8b8c1076","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/649958942ca6f96c8b8c1076/olfLIqNryaog1nAnQPwkN.jpeg","fullname":"Yudong Jin","name":"krahets","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":8,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.6968674659729004},"editors":["krahets"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/649958942ca6f96c8b8c1076/olfLIqNryaog1nAnQPwkN.jpeg"],"reactions":[],"isReport":false}},{"id":"6a87bba8a3a9beb9f84688c7","author":{"_id":"641c6e0692cd253029967564","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/641c6e0692cd253029967564/dlCk5kv9BlI8Gk06fsA1u.png","fullname":"Yujie","name":"Yujie0012","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":2,"isUserFollowing":false},"createdAt":"2026-08-21T02:44:56.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"👏","html":"👏\n","updatedAt":"2026-08-21T02:44:56.030Z","author":{"_id":"641c6e0692cd253029967564","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/641c6e0692cd253029967564/dlCk5kv9BlI8Gk06fsA1u.png","fullname":"Yujie","name":"Yujie0012","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":2,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"vi","probability":0.12120208144187927},"editors":["Yujie0012"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/641c6e0692cd253029967564/dlCk5kv9BlI8Gk06fsA1u.png"],"reactions":[{"reaction":"❤️","users":["krahets","Bannda"],"count":2}],"isReport":false}},{"id":"6a87d9e1b64531f0c763ccec","author":{"_id":"6a87d8c3170b15f3dcc49ec2","avatarUrl":"/avatars/7114d379fc94b2bb98185efd906adfda.svg","fullname":"Dong Qiang","name":"JohnQiangDong","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false},"createdAt":"2026-08-21T04:53:53.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"Absolute legend","html":"Absolute legend\n","updatedAt":"2026-08-21T04:53:53.668Z","author":{"_id":"6a87d8c3170b15f3dcc49ec2","avatarUrl":"/avatars/7114d379fc94b2bb98185efd906adfda.svg","fullname":"Dong Qiang","name":"JohnQiangDong","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"de","probability":0.18896260857582092},"editors":["JohnQiangDong"],"editorAvatarUrls":["/avatars/7114d379fc94b2bb98185efd906adfda.svg"],"reactions":[{"reaction":"🚀","users":["JohnQiangDong"],"count":1}],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2608.20335","authors":[{"_id":"6a87ac1289e517cbfd75dbad","name":"Yudong Jin","hidden":false},{"_id":"6a87ac1289e517cbfd75dbae","name":"Tao Xie","hidden":false},{"_id":"6a87ac1289e517cbfd75dbaf","name":"Qihang Zhang","hidden":false},{"_id":"6a87ac1289e517cbfd75dbb0","name":"Zehong Shen","hidden":false},{"_id":"6a87ac1289e517cbfd75dbb1","name":"Zhen Xu","hidden":false},{"_id":"6a87ac1289e517cbfd75dbb2","name":"Yujun Shen","hidden":false},{"_id":"6a87ac1289e517cbfd75dbb3","name":"Hujun Bao","hidden":false},{"_id":"6a87ac1289e517cbfd75dbb4","name":"Xiaowei Zhou","hidden":false},{"_id":"6a87ac1289e517cbfd75dbb5","name":"Yinghao Xu","hidden":false}],"mediaUrls":["https://cdn-uploads.huggingface.co/production/uploads/649958942ca6f96c8b8c1076/fKfNzciKY1L3I3XwM6SfY.mp4"],"publishedAt":"2026-08-20T00:00:00.000Z","submittedOnDailyAt":"2026-08-21T00:00:00.000Z","title":"4DAnyone: Create Anyone in 4D from a Casual Monocular Video","submittedOnDailyBy":{"_id":"649958942ca6f96c8b8c1076","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/649958942ca6f96c8b8c1076/olfLIqNryaog1nAnQPwkN.jpeg","isPro":false,"fullname":"Yudong Jin","user":"krahets","type":"user","name":"krahets"},"summary":"We present 4DAnyone, a framework for reconstructing 4D humans from an uncalibrated monocular video by generating reconstruction-grade multiview-consistent videos and lifting them into 4D Gaussian Splatting (4DGS). Existing camera-controlled video diffusion models synthesize plausible novel-view videos but fail to maintain consistency when scaled to the tens of target views required for 4DGS reconstruction. We identify this failure as a bounded-attention-context problem: when target views exceed the capacity of a single DiT forward pass, they must be split into groups, exposing two coupled bottlenecks. On the reference-context side, conditioning on all previously generated views grows as O(N), weakening cross-view appearance guidance. On the target-context side, disjoint groups cannot directly exchange information, causing global structural drift. 4DAnyone addresses both bottlenecks with two complementary designs: Reference Context Packing (RCP) compresses growing reference views into a fixed-length mixed-resolution context with O(1) reference-context complexity, while Target Context Routing (TCR) rotates target-view groupings during denoising to share context across groups at high-noise steps and stabilize details at low-noise steps. We further build the MVGameHuman dataset using our in-house game engine and combine it with light-stage and in-the-wild video datasets for training. Experiments on DNA-Rendering and DyMVHumans show that 4DAnyone outperforms prior methods in both novel-view video quality and downstream 4DGS reconstruction, with robust in-the-wild generalization. See our project page for video results and source code: https://4danyone.github.io.","upvotes":38,"discussionId":"6a87ac1389e517cbfd75dbb6","projectPage":"https://4danyone.github.io/","githubRepo":"https://github.com/ant-research/4DAnyone","githubRepoAddedBy":"user","ai_summary":"4DAnyone reconstructs 4D humans from monocular video by generating multiview-consistent videos and lifting them into 4D Gaussian Splatting, using reference and target context designs to overcome scaling bottlenecks.","ai_keywords":["4D Gaussian Splatting","video diffusion models","DiT","bounded-attention-context","Reference Context Packing","Target Context Routing","multiview-consistent video generation","4D human reconstruction"],"ai_summary_model":"thinkingmachines/Inkling-Small","githubStars":20,"organization":{"_id":"649824ccd52ece311d51c847","name":"AntResearch","fullname":"Ant Research","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/67440b1a180c3ec6c12c0b04/WeqzW2XDKtm1NGbttcZtI.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"649958942ca6f96c8b8c1076","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/649958942ca6f96c8b8c1076/olfLIqNryaog1nAnQPwkN.jpeg","isPro":false,"fullname":"Yudong Jin","user":"krahets","type":"user"},{"_id":"65fa86c8180d06d37c8f5c71","avatarUrl":"/avatars/21913d9a1d7585f73bd8a559520a3d8c.svg","isPro":false,"fullname":"Shuaiting Li","user":"list0830","type":"user"},{"_id":"6569ca717fd1d421381b1e5a","avatarUrl":"/avatars/080655b2497c2fa44a33da3ccfabb84b.svg","isPro":false,"fullname":"Jiaming Zhou","user":"Jiaming2472","type":"user"},{"_id":"641c6e0692cd253029967564","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/641c6e0692cd253029967564/dlCk5kv9BlI8Gk06fsA1u.png","isPro":false,"fullname":"Yujie","user":"Yujie0012","type":"user"},{"_id":"6654a66cac73f7759f764b4a","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/dl6Ajom1Xe9H067mRl-Qx.png","isPro":false,"fullname":"Jesse Senior","user":"jessesenior","type":"user"},{"_id":"62986ca2b58e71e2ac9b8f01","avatarUrl":"/avatars/83944db5f3dbb6f47c47c46fb2cb2849.svg","isPro":false,"fullname":"Sida Peng","user":"pengsida","type":"user"},{"_id":"648a7e986cc6b1f72e757a27","avatarUrl":"/avatars/6ac3f5ec7512143407f135534070b59f.svg","isPro":false,"fullname":"CatCoffee","user":"CatCoffee","type":"user"},{"_id":"653b560b025e5d01725dd1b0","avatarUrl":"/avatars/280b612c817caccc8086e2713398a705.svg","isPro":false,"fullname":"yyh","user":"yyh929","type":"user"},{"_id":"6870ba483391adb0d79c35f3","avatarUrl":"/avatars/90ff3c9b847180bb333e41f4f1f3d99a.svg","isPro":false,"fullname":"YihuiYang","user":"Yangyihui","type":"user"},{"_id":"676551601fde77e3bb968df9","avatarUrl":"/avatars/5ddd14228be7dcc0a443db8a984b7d19.svg","isPro":false,"fullname":"Taoming Liu","user":"xiaotao13","type":"user"},{"_id":"66b4839ee8632ddfcf836e14","avatarUrl":"/avatars/d635f2e09b3948f94cfbb15805bc91fc.svg","isPro":false,"fullname":"YunshengLuo","user":"YunshengLuo","type":"user"},{"_id":"67d8004877976629305e4b48","avatarUrl":"/avatars/2e883465efc2b7d1b7ab09ed041ec313.svg","isPro":false,"fullname":"lijiaxin","user":"lijiaxin2004","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":3,"organization":{"_id":"649824ccd52ece311d51c847","name":"AntResearch","fullname":"Ant Research","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/67440b1a180c3ec6c12c0b04/WeqzW2XDKtm1NGbttcZtI.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2608/2608.20335.md","query":{}}">

Papers

arxiv:2608.20335

4DAnyone: Create Anyone in 4D from a Casual Monocular Video

Published on Aug 20

· Submitted by

Yudong Jin on Aug 21

#3 Paper of the day

Ant Research

Upvote

Authors:

Abstract

4DAnyone reconstructs 4D humans from monocular video by generating multiview-consistent videos and lifting them into 4D Gaussian Splatting, using reference and target context designs to overcome scaling bottlenecks.

Generated by thinkingmachines/Inkling-Small

We present 4DAnyone, a framework for reconstructing 4D humans from an uncalibrated monocular video by generating reconstruction-grade multiview-consistent videos and lifting them into 4D Gaussian Splatting (4DGS). Existing camera-controlled video diffusion models synthesize plausible novel-view videos but fail to maintain consistency when scaled to the tens of target views required for 4DGS reconstruction. We identify this failure as a bounded-attention-context problem: when target views exceed the capacity of a single DiT forward pass, they must be split into groups, exposing two coupled bottlenecks. On the reference-context side, conditioning on all previously generated views grows as O(N), weakening cross-view appearance guidance. On the target-context side, disjoint groups cannot directly exchange information, causing global structural drift. 4DAnyone addresses both bottlenecks with two complementary designs: Reference Context Packing (RCP) compresses growing reference views into a fixed-length mixed-resolution context with O(1) reference-context complexity, while Target Context Routing (TCR) rotates target-view groupings during denoising to share context across groups at high-noise steps and stabilize details at low-noise steps. We further build the MVGameHuman dataset using our in-house game engine and combine it with light-stage and in-the-wild video datasets for training. Experiments on DNA-Rendering and DyMVHumans show that 4DAnyone outperforms prior methods in both novel-view video quality and downstream 4DGS reconstruction, with robust in-the-wild generalization. See our project page for video results and source code: https://4danyone.github.io.