
{"id":8858,"date":"2026-06-06T03:01:56","date_gmt":"2026-06-05T19:01:56","guid":{"rendered":"https:\/\/infernews.com\/blog\/official-implementation-of-loomvideo-unifying-multimodal-inputs-into-video-gener\/"},"modified":"2026-06-06T03:49:02","modified_gmt":"2026-06-05T19:49:02","slug":"loomvideo","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/loomvideo\/","title":{"rendered":"LoomVideo\uff1a\u963f\u91cc\u5df4\u5df4 5B \u6a21\u578b\u6311\u6230\u7d71\u4e00\u5f71\u7247\u751f\u6210"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/06\/pasted-b42add55c8a9.jpg\" alt=\"architecture\"\/><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">LoomVideo \u7531\u5317\u4eac\u5927\u5b78\u8207\u963f\u91cc\u5df4\u5df4\u806f\u5408\u767c\u5e03\uff0c\u4e3b\u6253\u300c\u7d71\u4e00\u591a\u6a21\u614b\u8f38\u5165\u7684\u5f71\u7247\u751f\u6210\u8207\u7de8\u8f2f\u300d\uff0c\u628a\u6587\u5b57\u3001\u5f71\u7247\u3001\u5716\u7247\u7b49\u591a\u7a2e\u8f38\u5165\u6574\u5408\u5230\u540c\u4e00\u500b\u6a21\u578b\u3002\u50b3\u7d71\u7684\u7d71\u4e00\u5f71\u7247\u6a21\u578b\u52d5\u8f12\u8d85\u904e 13B \u53c3\u6578\uff0c\u4e14\u70ba\u4e86\u52a0\u5165\u4f86\u6e90\u5f71\u7247\u689d\u4ef6\uff0c\u5f80\u5f80\u8981\u628a\u6240\u6709 token \u63a5\u5728\u4e00\u8d77\uff0c\u5c0e\u81f4\u5e8f\u5217\u9577\u5ea6\u7ffb\u500d\u3001self-attention \u6210\u672c\u66b4\u589e\u56db\u500d\u3002LoomVideo \u7684\u6838\u5fc3\u5b9a\u4f4d\uff0c\u5c31\u662f\u7528\u66f4\u5c0f\u3001\u66f4\u5feb\u7684\u8a2d\u8a08\uff0c\u9054\u5230\u540c\u7b49\u751a\u81f3\u66f4\u597d\u7684\u6548\u679c\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u9019\u500b\u9805\u76ee\u57fa\u65bc MLLM\uff08\u591a\u6a21\u614b\u5927\u8a9e\u8a00\u6a21\u578b\uff09\u52a0 DiT\uff08Diffusion Transformer\uff09\u7684\u7d44\u5408\uff0c\u4e26\u63d0\u51fa\u4e09\u500b\u95dc\u9375\u8a2d\u8a08\uff1a<strong>Deepstack Injection<\/strong> \u5f9e MLLM \u6bcf\u4e00\u5c64\u62bd\u53d6\u7279\u5fb5\uff0c\u518d\u900f\u904e cross-attention \u6ce8\u5165\u5c0d\u61c9\u7684 DiT \u5c64\uff1b<strong>Scale-and-Add Conditioning<\/strong> \u628a\u4e7e\u6de8\u7684\u4f86\u6e90\u5f71\u7247\u6f5b\u5728\u8b8a\u6578\u6309\u6642\u9593\u6b65\u9577\u7e2e\u653e\u5f8c\u76f4\u63a5\u52a0\u5230\u96dc\u8a0a\u76ee\u6a19\u4e0a\uff0c\u514d\u53bb token \u62fc\u63a5\u7684\u984d\u5916\u8ca0\u64d4\uff1b<strong>Negative Temporal RoPE<\/strong> \u70ba\u53c3\u8003\u5716\u7247\u6307\u5b9a\u8ca0\u7684\u6642\u9593\u7d22\u5f15\uff0c\u8b93\u591a\u5716\u7247\u689d\u4ef6\u53ef\u4ee5\u7121\u7e2b\u6574\u5408\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">LoomVideo \u76ee\u524d\u652f\u63f4\u56db\u7a2e\u4efb\u52d9\uff1a\u6587\u751f\u5f71\u7247\u3001\u7d14\u6587\u5b57\u6307\u4ee4\u7de8\u8f2f\u3001\u5f71\u7247\u52a0\u5716\u7247\u52a0\u6587\u5b57\u7684\u6307\u4ee4\u7de8\u8f2f\uff0c\u4ee5\u53ca\u591a\u5f35\u53c3\u8003\u5716\u7684\u689d\u4ef6\u751f\u6210\uff0c\u5168\u90e8\u7531\u540c\u4e00\u500b 5B \u6a21\u578b\u8655\u7406\u3002\u8ad6\u6587\u5831\u544a\u5728\u591a\u9805\u57fa\u6e96\u4e0a\u53d6\u5f97\u9818\u5148\u6216\u5177\u7af6\u722d\u529b\u7684\u8868\u73fe\uff0c\u4e26\u5ba3\u7a31\u6bd4\u540c\u7d1a\u6a21\u578b\u5feb\u81f3\u5c11 <strong>5.41 \u500d<\/strong>\u3002\u5c0d\u65bc\u96fb\u5b50\u5546\u52d9\u8207\u6642\u5c1a\u5834\u666f\u7684\u5f71\u7247\u751f\u6210\uff0c\u8ad6\u6587\u4e5f\u5c55\u793a\u4e86\u91dd\u5c0d\u6027\u7684\u512a\u52e2\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u9019\u500b\u9805\u76ee\u9069\u5408\u95dc\u6ce8\u5f71\u7247\u751f\u6210\u6548\u7387\u7684\u7814\u7a76\u8005\u3001\u5f71\u50cf\u5275\u4f5c\u5de5\u5177\u958b\u767c\u8005\uff0c\u4ee5\u53ca\u9700\u8981\u5feb\u901f\u751f\u6210\u77ed\u5f71\u7247\u5167\u5bb9\u7684\u5718\u968a\u3002\u76ee\u524d\u6a21\u578b\u6b0a\u91cd\u5df2\u516c\u958b\u5728 Hugging Face \u7684 <code>MSALab\/LoomVideo<\/code>\uff0c\u7a0b\u5f0f\u78bc\u4ea6\u540c\u6b65\u91cb\u51fa\uff0c\u6709\u8208\u8da3\u7684\u8b80\u8005\u53ef\u4ee5\u76f4\u63a5\u5230 GitHub \u8207 Hugging Face \u53d6\u5f97\u8cc7\u6e90\u4e26\u6e2c\u8a66\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u91cd\u9ede\u6458\u8981<\/strong>\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>5B \u53c3\u6578\u7684\u7d71\u4e00\u5f71\u7247\u751f\u6210\u8207\u7de8\u8f2f\u6a21\u578b\uff0c\u5b9a\u4f4d\u6bd4 13B+ \u540c\u985e\u66f4\u8f15\u91cf\u3002<\/li>\n\n\n\n<li>\u4ee5 MLLM + DiT \u67b6\u69cb\u70ba\u57fa\u790e\uff0c\u4e26\u63d0\u51fa Deepstack Injection\u3001Scale-and-Add Conditioning \u8207 Negative Temporal RoPE \u4e09\u5927\u8a2d\u8a08\u3002<\/li>\n\n\n\n<li>\u652f\u63f4\u6587\u751f\u5f71\u7247\u3001\u6587\u5b57\u6307\u4ee4\u7de8\u8f2f\u3001\u5f71\u7247\u52a0\u5716\u7247\u6587\u5b57\u7de8\u8f2f\uff0c\u4ee5\u53ca\u591a\u5716\u7247\u689d\u4ef6\u751f\u6210\u56db\u7a2e\u4efb\u52d9\u3002<\/li>\n\n\n\n<li>\u8ad6\u6587\u8072\u7a31\u6bd4\u540c\u7d1a\u6a21\u578b\u5feb\u81f3\u5c11 5.41 \u500d\uff0c\u4e26\u5728\u96fb\u5546\u8207\u6642\u5c1a\u5834\u666f\u8868\u73fe\u7a81\u51fa\u3002<\/li>\n\n\n\n<li>\u6a21\u578b\u8207\u7a0b\u5f0f\u78bc\u5df2\u516c\u958b\uff0c\u65b9\u4fbf\u7814\u7a76\u8005\u8207\u958b\u767c\u8005\u5feb\u901f\u8a66\u7528\u8207\u6539\u9032\u3002<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>GitHub\uff1a<\/strong> <a href=\"https:\/\/github.com\/MSALab-PKU\/LoomVideo\" rel=\"noopener noreferrer\">https:\/\/github.com\/MSALab-PKU\/LoomVideo<\/a><\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u9805\u76ee\uff1a<\/strong> <a href=\"https:\/\/msalab-pku.github.io\/projects\/LoomVideo\/index.html\" rel=\"noopener noreferrer\">https:\/\/msalab-pku.github.io\/projects\/LoomVideo\/index.html<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u5317\u4eac\u5927\u5b78\u8207\u963f\u91cc\u5408\u4f5c\u63a8\u51fa\u7684 LoomVideo \u4ee5 5B \u53c3\u6578\u7684\u8f15\u91cf\u67b6\u69cb\uff0c\u89e3\u6c7a\u7d71\u4e00\u5f71\u7247\u751f\u6210\u6210\u672c\u904e\u9ad8\u7684\u75db\u9ede\uff0c\u4e26\u5e36\u4f86\u986f\u8457\u7684\u901f\u5ea6\u63d0\u5347\u3002<\/p>\n","protected":false},"author":8,"featured_media":8857,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"ai_generated_summary":"","wpai_meta_description":"","footnotes":""},"categories":[133,171,120,76,126,149,141,195],"tags":[],"class_list":["post-8858","post","type-post","status-publish","format-standard","hentry","category-133","category-171","category-120","category-76","category-126","category-149","category-141","category-195"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/8858","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/users\/8"}],"replies":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/comments?post=8858"}],"version-history":[{"count":2,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/8858\/revisions"}],"predecessor-version":[{"id":8863,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/8858\/revisions\/8863"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/8857"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=8858"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=8858"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=8858"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}