We present <strong>SolarWM</strong>, a fully open foundation for building interactive video<br>world models from data preparation through scalable training and long-horizon<br>inference.</p>\n<ul>\n<li><strong>Open, reconfigurable data infrastructure.</strong> SolarWM converts 1.43 million<br>canonical clips from 14 datasets into a unified, frame-aligned contract for<br>observations, metric camera geometry, captions, quality metadata, selection,<br>and provenance. Source processing is decoupled from training-mixture design.</li>\n<li><strong>A scalable, backbone-native model family.</strong> One framework supports four<br>5B–33B models across Wan2.2, LTX-2.5, and MiniMax-H3 while preserving each<br>backbone's native representation and objective.</li>\n<li><strong>A simple three-stage training recipe.</strong> Bidirectional adaptation,<br>teacher-forced autoregressive initialization, and distribution matching<br>distillation form a shared route across heterogeneous video backbones,<br>without specialized ODE or consistency-distillation initialization.</li>\n<li><strong>Long-horizon interaction from short training clips.</strong> After training only<br>on 5-second sequences, the resulting causal models support real-time<br>interaction with rollouts spanning minutes to hours, without long-sequence<br>fine-tuning or attention-sink mechanisms.</li>\n</ul>\n","updatedAt":"2026-09-03T02:04:16.609Z","author":{"_id":"68961e2a83f79bb28f78baa3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/bZmicB1lDqFjIVbpsg_8c.png","fullname":"junchao-cuhk","name":"junchao-cuhk","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":1,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8493039608001709},"editors":["junchao-cuhk"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/bZmicB1lDqFjIVbpsg_8c.png"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2609.02886","authors":[{"_id":"6a98d2befea818274321fe2a","name":"Junchao Huang","hidden":false},{"_id":"6a98d2befea818274321fe2b","user":{"_id":"623461fccd8a0462e55b3666","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1647600114080-noauth.jpeg","isPro":true,"fullname":"Guian Fang","user":"Enderfga","type":"user","name":"Enderfga"},"name":"Guian Fang","status":"claimed_verified","statusLastChangedAt":"2026-09-03T08:17:10.448Z","hidden":false},{"_id":"6a98d2befea818274321fe2c","name":"Shengju Qian","hidden":false},{"_id":"6a98d2befea818274321fe2d","user":{"_id":"6428fd124fe87caede856311","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/5OrNPwZkxu3Dm1IInCxML.jpeg","isPro":false,"fullname":"Xianghao Kong","user":"refkxh","type":"user","name":"refkxh"},"name":"Xianghao Kong","status":"claimed_verified","statusLastChangedAt":"2026-09-03T08:17:12.077Z","hidden":false},{"_id":"6a98d2befea818274321fe2e","user":{"_id":"67f87bc19d597ac661a75b68","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/67f87bc19d597ac661a75b68/ARLLbu1CJ5mCQu6ptyfmG.jpeg","isPro":false,"fullname":"Zhuoran Zhao","user":"Alicezrzhao","type":"user","name":"Alicezrzhao"},"name":"Zhuoran Zhao","status":"claimed_verified","statusLastChangedAt":"2026-09-03T08:45:04.309Z","hidden":false},{"_id":"6a98d2befea818274321fe2f","name":"Wei Huang","hidden":false},{"_id":"6a98d2befea818274321fe30","name":"Yihua Du","hidden":false},{"_id":"6a98d2befea818274321fe31","name":"Zixin Zhang","hidden":false},{"_id":"6a98d2befea818274321fe32","name":"Justin Cui","hidden":false},{"_id":"6a98d2befea818274321fe33","name":"Yuchao Gu","hidden":false},{"_id":"6a98d2befea818274321fe34","name":"Yukang Chen","hidden":false},{"_id":"6a98d2befea818274321fe35","name":"Xinting Hu","hidden":false},{"_id":"6a98d2befea818274321fe36","name":"Tianyu He","hidden":false},{"_id":"6a98d2befea818274321fe37","name":"Shaoshuai Shi","hidden":false},{"_id":"6a98d2befea818274321fe38","name":"Zhuotao Tian","hidden":false},{"_id":"6a98d2befea818274321fe39","name":"Xin Wang","hidden":false},{"_id":"6a98d2befea818274321fe3a","name":"Mike Zheng Shou","hidden":false},{"_id":"6a98d2befea818274321fe3b","name":"Li Jiang","hidden":false}],"publishedAt":"2026-09-02T00:00:00.000Z","submittedOnDailyAt":"2026-09-03T00:00:00.000Z","title":"SolarWM: Open Data and Scalable Training for Long-Horizon Video World Models","submittedOnDailyBy":{"_id":"68961e2a83f79bb28f78baa3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/bZmicB1lDqFjIVbpsg_8c.png","isPro":false,"fullname":"junchao-cuhk","user":"junchao-cuhk","type":"user","name":"junchao-cuhk"},"summary":"We introduce SolarWM, a fully open foundation for building interactive video world models from data preparation through long-horizon inference. Training across heterogeneous data sources and video backbones is challenging: datasets differ in temporal scale, camera geometry, visual quality, motion, and captioning styles, while video generators use distinct representations and architectures. Naive data mixing and model-specific implementations therefore produce inconsistent supervision and make results difficult to reproduce and compare. SolarWM addresses this coupling with a reconfigurable multi-source data engine and a backbone-native adaptation framework. The engine converts 1.43 million canonical clips from 10 datasets into a unified, frame-aligned contract covering visual observations, metric camera geometry, captions, quality metadata, selection decisions, and provenance, while decoupling source processing from mixture construction. Under shared camera-conditioning, training, and inference interfaces, we instantiate four 5B--33B models based on Wan2.2, LTX-2.5, and MiniMax-H3 while preserving their native representations and objectives. A unified three-stage recipe combines bidirectional adaptation, teacher-forced autoregressive initialization, and distribution matching distillation. The resulting causal models enable real-time interaction over rollouts ranging from minutes to hours after being trained on only 5s sequences. By releasing the resulting data, pipeline, recipes, weights, and framework, SolarWM provides a reproducible and extensible foundation for interactive world-model research.","upvotes":105,"discussionId":"6a98d2bffea818274321fe3c","projectPage":"https://junchao-cs.github.io/SolarWM-Web/","githubRepo":"https://github.com/Junchao-cs/SolarWM","githubRepoAddedBy":"user","ai_summary":"SolarWM provides an open framework and unified training recipe for building interactive video world models across diverse data sources and generator backbones, enabling long-horizon real-time rollouts.","ai_keywords":["video world models","multi-source data engine","backbone-native adaptation","camera-conditioning","autoregressive initialization","distribution matching distillation","causal models"],"ai_summary_model":"thinkingmachines/Inkling-Small","organization":{"_id":"6223644d0129f2097d69a407","name":"CUHKSZ","fullname":"Chinese University of Hong Kong, Shenzhen","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/1646486592158-6108ae87823007eaf0c7bd1e.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"68961e2a83f79bb28f78baa3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/bZmicB1lDqFjIVbpsg_8c.png","isPro":false,"fullname":"junchao-cuhk","user":"junchao-cuhk","type":"user"},{"_id":"6428fd124fe87caede856311","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/5OrNPwZkxu3Dm1IInCxML.jpeg","isPro":false,"fullname":"Xianghao Kong","user":"refkxh","type":"user"},{"_id":"67f87bc19d597ac661a75b68","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/67f87bc19d597ac661a75b68/ARLLbu1CJ5mCQu6ptyfmG.jpeg","isPro":false,"fullname":"Zhuoran Zhao","user":"Alicezrzhao","type":"user"},{"_id":"65214c46f6ceb915cc790275","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/65214c46f6ceb915cc790275/Jb-iNYkDc9g5nODi0stGE.png","isPro":false,"fullname":"Yihua Du","user":"Duyh","type":"user"},{"_id":"662d968eac05b4f7c280d6f9","avatarUrl":"/avatars/0fb03561fff425fddfa48e1ebde6325b.svg","isPro":false,"fullname":"YaoYang Liu","user":"LazySheeep","type":"user"},{"_id":"623461fccd8a0462e55b3666","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1647600114080-noauth.jpeg","isPro":true,"fullname":"Guian Fang","user":"Enderfga","type":"user"},{"_id":"6a81771e1806973d0d3c3491","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6a81771e1806973d0d3c3491/AqmLPhfFmCyeidzzKFZ9x.jpeg","isPro":false,"fullname":"Winnie Saputra","user":"winniesaputra","type":"user"},{"_id":"6039478ab3ecf716b1a5fd4d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6039478ab3ecf716b1a5fd4d/_Thy4E7taiSYBLKxEKJbT.jpeg","isPro":true,"fullname":"taesiri","user":"taesiri","type":"user"},{"_id":"68f792b148e9f4a46eca68bf","avatarUrl":"/avatars/a6442a2777a4b7e78f01168b80847446.svg","isPro":false,"fullname":"Fukang Wen","user":"smallkang2025","type":"user"},{"_id":"6a05607454ba3c3be9fe749e","avatarUrl":"/avatars/5fd59a7c2c2215d86e4be9de413df2c6.svg","isPro":false,"fullname":"Dai Jiahui","user":"jiahuiiiiiiiii","type":"user"},{"_id":"6a0c1e01531ade4e2540483f","avatarUrl":"/avatars/9765a3a3f9ad25cbf1e19ab589379bda.svg","isPro":false,"fullname":"yuxiang feng","user":"yuxiang921","type":"user"},{"_id":"6411c801e872ae3fb1e2c96e","avatarUrl":"/avatars/f8898dc13d700e545eedbbfab1c18353.svg","isPro":true,"fullname":"Franklin","user":"Franklinzhang","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":2,"organization":{"_id":"6223644d0129f2097d69a407","name":"CUHKSZ","fullname":"Chinese University of Hong Kong, Shenzhen","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/1646486592158-6108ae87823007eaf0c7bd1e.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2609/2609.02886.md","query":{}}">

Get this paper in your agent:

Don't have the latest CLI?

Cite arxiv.org/abs/2609.02886 in a Space README.md to link it from this page.