[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-new-method-trains-ai-video-generators-without-extra-models":10,"sections":40},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news, decoded.","editor@therevision.news",{"gaMeasurementId":8,"adsenseClientId":9},"G-ZW2MV82GYR","ca-pub-8533917693782264",{"article":11},{"id":12,"slug":13,"title":14,"dek":15,"body_md":16,"tags_json":17,"published_at":18,"created_at":19,"updated_at":20,"status":21,"review_note":22,"review_notes":23,"image_url":22,"persona_id":22,"persona_name":22,"section":30,"tags":31,"sources":35,"feedback":39,"feedback_at":22,"cost_usd":39,"total_tokens":39},8736,"new-method-trains-ai-video-generators-without-extra-models","New Method Trains AI Video Generators Without Extra Models","A new paper's Elastic Forcing method trains autoregressive video AI from real footage, dropping the extra teacher and fake-score networks most systems require.","A new training method lets AI video generators learn straight from real footage, cutting out the extra teacher models that usually drive up training cost.\n\nThe technique, called Elastic Forcing, comes from a paper titled Elastic Forcing: From Scores to Samples for Autoregressive Video Generation (arXiv:2609.35491). Most few-step autoregressive video generators rely on Distribution Matching Distillation, which needs a bidirectional diffusion teacher model plus a second fake-score model running online during training. Elastic Forcing skips both: it learns the rollout distribution directly from reference videos by minimizing the gap between generated and real video statistics in a frozen, pre-trained video representation space, using a hybrid Nystrom-Monte Carlo estimator to keep that comparison accurate without blowing up compute. Using the same architecture and starting point as the earlier Self-Forcing system, the 1.3-billion-parameter model raised its VBench quality score from 83.80 to 84.64 while still generating at 17 frames per second, and removing the extra score models let the team train a larger 14-billion-parameter version on just eight H200 GPUs.\n\nFewer moving parts during training matters because those auxiliary score models are expensive to run and cap how big a model you can practically train. Cutting them frees up the same compute budget for a model more than ten times larger, and the underlying idea, learning straight from reference videos, points toward teaching video AI new visual styles or concepts without building a matching diffusion teacher for each one.\n\nA less-than-one-point bump on a single benchmark won't turn heads, but trimming teacher models out of the training loop is the sort of unglamorous efficiency gain that tends to matter more once everyone else has to catch up.","[\"ai\",\"video-generation\",\"machine-learning\",\"research\"]","2026-09-30T04:00:00.000Z","2026-09-30T22:45:35.482Z","2026-09-30T22:45:42.077Z","published",null,[24],{"id":25,"reviewer":26,"round":27,"reason":28,"status":29},"editor-r1","editor",1,"Name and cite the source paper (title 'Elastic Forcing: From Scores to Samples for Autoregressive Video Generation,' arXiv ID, and authors\u002Finstitution if known) instead of referring to it only as 'this paper,' so readers can verify the claims.","resolved","ai",[30,32,33,34],"video-generation","machine-learning","research",[36],{"name":37,"url":38},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.35491",0,{"sections":41},[42,46,51,56,61,65,69,73,78,82,87,92,97,102],{"name":43,"slug":30,"count":44,"latest_published_at":45},"AI",5214,"2026-09-30T13:00:00.000Z",{"name":47,"slug":48,"count":49,"latest_published_at":50},"Security","security",793,"2026-09-30T12:55:00.000Z",{"name":52,"slug":53,"count":54,"latest_published_at":55},"Policy","policy",419,"2026-09-30T12:24:32.000Z",{"name":57,"slug":58,"count":59,"latest_published_at":60},"Deals","deals",291,"2026-09-30T10:38:22.000Z",{"name":62,"slug":63,"count":64,"latest_published_at":45},"Hardware","hardware",196,{"name":66,"slug":67,"count":68,"latest_published_at":18},"Science","science",155,{"name":70,"slug":71,"count":72,"latest_published_at":45},"Consumer Tech","consumer-tech",144,{"name":74,"slug":75,"count":76,"latest_published_at":77},"Dev Tools","dev-tools",91,"2026-09-30T12:58:00.000Z",{"name":79,"slug":80,"count":76,"latest_published_at":81},"Software","software","2026-09-25T20:55:00.000Z",{"name":83,"slug":84,"count":85,"latest_published_at":86},"Startups","startups",83,"2026-09-29T21:51:36.000Z",{"name":88,"slug":89,"count":90,"latest_published_at":91},"General","general",49,"2026-09-28T16:44:57.000Z",{"name":93,"slug":94,"count":95,"latest_published_at":96},"Gaming","gaming",48,"2026-09-25T18:35:21.000Z",{"name":98,"slug":99,"count":100,"latest_published_at":101},"Reviews","reviews",31,"2026-09-28T14:31:34.000Z",{"name":103,"slug":104,"count":105,"latest_published_at":106},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]