{"id":28671,"date":"2026-01-08T10:49:19","date_gmt":"2026-01-08T02:49:19","guid":{"rendered":"https:\/\/aif.amtbbs.org\/?p=28671"},"modified":"2026-01-08T10:49:19","modified_gmt":"2026-01-08T02:49:19","slug":"multitalk%ef%bc%9a%e5%a4%9a%e8%a7%92%e8%89%b2%e5%af%b9%e8%af%9d%e7%94%9f%e6%88%90sota%e6%a8%a1%e5%9e%8b%ef%bc%8c%e8%af%ad%e9%9f%b3-%e8%a7%86%e8%a7%89%e5%af%b9%e9%bd%90%e7%b2%be%e5%ba%a6%e8%be%be98-7","status":"publish","type":"post","link":"https:\/\/aif.amtbbs.org\/index.php\/2026\/01\/08\/28671\/","title":{"rendered":"MultiTalk\uff1a\u591a\u89d2\u8272\u5bf9\u8bdd\u751f\u6210SOTA\u6a21\u578b\uff0c\u8bed\u97f3-\u89c6\u89c9\u5bf9\u9f50\u7cbe\u5ea6\u8fbe98.7%\uff01"},"content":{"rendered":"
MultiTalk\u4ee5DiT\uff08Diffusion-in-Transformer\uff09\u4e3a\u57fa\u7840\u7684\u89c6\u9891\u6269\u6563\u6a21\u578b\u4f5c\u4e3a\u5176\u6838\u5fc3\u9aa8\u67b6\u3002<\/div>\n
\n
\n

\u7531\u4e2d\u5c71\u5927\u5b66\u3001\u7f8e\u56e2\u3001\u6e2f\u79d1\u5927\u5f00\u6e90\u7684\u00a0MultiTalk \u53ef\u5b9e\u73b0\u591a\u865a\u62df\u4eba\u5bf9\u8bdd\u89c6\u9891\u751f\u6210\u3002\u5728\u8bed\u97f3\u4e0e\u5634\u5f62\u540c\u6b65\u65b9\u9762\u8fbe\u5230\u4e86SOTA\u6027\u80fd\uff0c\u5e76\u652f\u6301\u901a\u8fc7prompt\u5b9e\u73b0\u4eba\u7269\u3001\u7269\u4f53\u4e0e\u573a\u666f\u7684\u4ea4\u4e92\u3002<\/p>\n

<\/p>\n

<\/p>\n

<\/p>\n

<\/p>\n

\u76f8\u5173\u94fe\u63a5<\/h4>\n
    \n
  • \u4e3b\u9875\uff1ahttps:\/\/meigen-ai.github.io\/multi-talk\/<\/li>\n
  • \u4ee3\u7801\uff1ahttps:\/\/github.com\/MeiGen-AI\/MultiTalk<\/li>\n
  • \u8bba\u6587\uff1ahttps:\/\/arxiv.org\/abs\/2505.22647<\/li>\n<\/ul>\n

    \u8bba\u6587\u4ecb\u7ecd<\/h4>\n

    <\/p>\n

    \u8fd1\u5e74\u6765\uff0c\u97f3\u9891\u9a71\u52a8\u7684\u4eba\u4f53\u52a8\u753b\u6280\u672f\u98de\u901f\u53d1\u5c55\uff0c\u4ece\u903c\u771f\u7684\u8bf4\u8bdd\u5934\u90e8\uff08Talking Head\uff09\u5230\u5168\u8eab\u52a8\u4f5c\u540c\u6b65\uff08Talking Body\uff09\uff0c\u5df2\u80fd\u751f\u6210\u9ad8\u81ea\u7136\u5ea6\u7684\u5355\u4eba\u7269\u89c6\u9891\u3002\u7136\u800c\uff0c\u73b0\u6709\u6280\u672f\u5927\u591a\u5c40\u9650\u4e8e\u5355\u4eba\u573a\u666f\uff0c\u9762\u5bf9\u591a\u4eba\u5bf9\u8bdd\u89c6\u9891\u751f\u6210\u65f6\u9762\u4e34\u4e09\u5927\u6311\u6218\uff1a<\/p>\n

      \n
    1. \u591a\u97f3\u9891\u6d41\u8f93\u5165\u9002\u914d\uff1a\u5982\u4f55\u533a\u5206\u5e76\u7ed1\u5b9a\u4e0d\u540c\u4eba\u7269\u7684\u97f3\u9891\u4fe1\u53f7\uff1f<\/li>\n
    2. \u52a8\u6001\u4eba\u7269\u5b9a\u4f4d\uff1a\u5f53\u4eba\u7269\u5728\u753b\u9762\u4e2d\u79fb\u52a8\u65f6\uff0c\u5982\u4f55\u7cbe\u51c6\u5b9a\u4f4d\u5176\u8fd0\u52a8\u533a\u57df\uff1f<\/li>\n
    3. \u6307\u4ee4\u9075\u5faa\u80fd\u529b\uff1a\u5982\u4f55\u8ba9\u751f\u6210\u7684\u89c6\u9891\u4e25\u683c\u9075\u5faa\u6587\u672c\u63cf\u8ff0\u7684\u590d\u6742\u52a8\u4f5c\uff08\u5982\u5927\u5e45\u80a2\u4f53\u52a8\u4f5c\uff09\uff1f<\/li>\n<\/ol>\n

      \u65b9\u6cd5\u6982\u8ff0<\/h4>\n

      <\/p>\n

      MultiTalk\u4ee5DiT\uff08Diffusion-in-Transformer\uff09\u4e3a\u57fa\u7840\u7684\u89c6\u9891\u6269\u6563\u6a21\u578b\u4f5c\u4e3a\u5176\u6838\u5fc3\u9aa8\u67b6\u3002\u57fa\u7840\u7684\u56fe\u50cf\u5230\u89c6\u9891\uff08I2V\uff09\u6269\u6563\u6a21\u578b\u901a\u5e38\u4e0d\u539f\u751f\u652f\u6301\u97f3\u9891\u8f93\u5165\u3002\u4e3a\u4e86\u8ba9\u6a21\u578b\u80fd\u591f\u201c\u8bf4\u8bdd\u201d\uff0cMultiTalk\u5728\u6bcf\u4e2aDiT\u5757\u7684\u6587\u672c\u4ea4\u53c9\u6ce8\u610f\u529b\u5c42\u4e4b\u540e\uff0c\u6dfb\u52a0\u4e86\u65b0\u7684\u5c42\uff0c\u8fd9\u4e9b\u5c42\u5305\u542b\u5c42\u5f52\u4e00\u5316\u548c\u97f3\u9891\u4ea4\u53c9\u6ce8\u610f\u529b\u673a\u5236\uff0c\u4e13\u95e8\u7528\u4e8e\u5904\u7406\u548c\u6574\u5408\u97f3\u9891\u6761\u4ef6\u3002<\/p>\n

      \u4f20\u7edf\u65b9\u6cd5\u76f4\u63a5\u878d\u5408\u591a\u97f3\u9891\u6d41\u4f1a\u5bfc\u81f4\u4eba\u7269\u4e0e\u97f3\u9891\u9519\u914d\u3002MultiTalk\u63d0\u51faLabel Rotary Position Embedding\uff08L-RoPE\uff09\uff0c\u901a\u8fc7\u4ee5\u4e0b\u6b65\u9aa4\u5b9e\u73b0\u7cbe\u51c6\u7ed1\u5b9a\uff1a<\/p>\n

        \n
      • \u6b65\u9aa41\uff1a\u81ea\u9002\u5e94\u4eba\u7269\u5b9a\u4f4d\u5229\u7528\u53c2\u8003\u56fe\u50cf\u7684\u81ea\u6ce8\u610f\u529b\u56fe\uff08Self-Attention Map\uff09\uff0c\u8ba1\u7b97\u4eba\u7269\u533a\u57df\u4e0e\u80cc\u666f\u7684\u76f8\u4f3c\u5ea6\u77e9\u9635\uff0c\u52a8\u6001\u5206\u5272\u89c6\u9891\u6f5c\u5728\u7279\u5f81\uff08Video Latents\uff09\u4e3a\u4e0d\u540c\u4eba\u7269\u533a\u57df\u3002<\/li>\n
      • \u6b65\u9aa42\uff1a\u6807\u7b7e\u5206\u914d\u4e0e\u65cb\u8f6c\u7f16\u7801\u4e3a\u6bcf\u4e2a\u8bf4\u8bdd\u4eba\u5206\u914d\u72ec\u7acb\u7684\u6570\u503c\u8303\u56f4\u6807\u7b7e\uff08\u5982Person1\uff1a0-4\uff0cPerson2\uff1a20-24\uff09\uff0c\u5e76\u901a\u8fc7\u65cb\u8f6c\u4f4d\u7f6e\u7f16\u7801\uff08RoPE\uff09\u5c06\u6807\u7b7e\u6620\u5c04\u5230\u97f3\u9891\u4ea4\u53c9\u6ce8\u610f\u529b\u5c42\u3002\u76f8\u540c\u6807\u7b7e\u7684\u97f3\u9891\u4e0e\u89c6\u9891\u533a\u57df\u4f1a\u88ab\u6fc0\u6d3b\uff0c\u4ece\u800c\u7ed1\u5b9a\u97f3\u9891\u4e0e\u4eba\u7269\u7684\u5507\u90e8\u52a8\u4f5c\u3002<\/li>\n<\/ul>\n

        <\/p>\n

        \u5b9e\u9a8c\u7ed3\u679c<\/h4>\n

        <\/p>\n

        \u7ed3\u8bba<\/h4>\n

        MultiTalk\u63d0\u51fa\u4e00\u79cd\u97f3\u9891\u9a71\u52a8\u591a\u4eba\u7269\u5bf9\u8bdd\u89c6\u9891\u751f\u6210\u65b9\u6848\uff0c\u5176\u6838\u5fc3\u7a81\u7834\u5728\u4e8e\u5176\u521b\u65b0\u7684L-ROPE\u65b9\u6cd5\uff0c\u5b83\u901a\u8fc7\u7ed3\u5408\u81ea\u9002\u5e94\u4eba\u7269\u5b9a\u4f4d\u548c\u5e26\u6709\u7c7b\u522b\u4fe1\u606f\u7684\u6807\u7b7e\u7f16\u7801\uff0c\u6709\u6548\u89e3\u51b3\u4e86\u591a\u6d41\u97f3\u9891\u7684\u6ce8\u5165\u548c\u4eba\u7269\u7ed1\u5b9a\u8fd9\u4e00\u96be\u9898\u3002\u6b64\u5916\uff0c\u5176\u7cbe\u5fc3\u8bbe\u8ba1\u7684\u90e8\u5206\u53c2\u6570\u8bad\u7ec3\u548c\u591a\u4efb\u52a1\u8bad\u7ec3\u7b56\u7565\uff0c\u786e\u4fdd\u4e86\u6a21\u578b\u5728\u6709\u9650\u8d44\u6e90\u4e0b\u4f9d\u7136\u80fd\u591f\u4fdd\u6301\u5f3a\u5927\u7684\u6307\u4ee4\u9075\u5faa\u80fd\u529b\u548c\u9ad8\u8d28\u91cf\u7684\u89c6\u89c9\u8f93\u51fa\u3002MultiTalk\u9996\u6b21\u5c06\u8bed\u97f3\u9a71\u52a8\u7684\u52a8\u753b\u4ece\u5355\u4eba\u63a8\u5411\u591a\u4eba\u573a\u666f\uff0c\u4e3a\u865a\u62df\u4e3b\u64ad\u3001\u5f71\u89c6\u5236\u4f5c\u7b49\u9886\u57df\u63d0\u4f9b\u4e86\u5f3a\u6709\u529b\u5de5\u5177\u3002<\/p>\n

        \u6587\u7ae0\u6765\u81ea\uff1a51CTO<\/p>\n<\/div>\n<\/div>\n

        <\/div>\n

        <\/g><\/svg><\/i> \"Loading\"<\/p>\n

        <\/div>\n","protected":false},"excerpt":{"rendered":"

        MultiTalk\u4ee5DiT\uff08Diffusion-in-Transformer\uff09\u4e3a\u57fa\u7840\u7684\u89c6\u9891\u6269\u6563\u6a21\u578b\u4f5c\u4e3a\u5176\u6838\u5fc3\u9aa8 […]<\/p>\n

        <\/div>\n

        <\/g><\/svg><\/i> \"Loading\"<\/p>\n

        <\/div>\n","protected":false},"author":56,"featured_media":28672,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[23,20,80],"tags":[2068],"class_list":["post-28671","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-23","category-20","category-80","tag-multitalksota-98-7"],"_links":{"self":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/28671","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/users\/56"}],"replies":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/comments?post=28671"}],"version-history":[{"count":1,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/28671\/revisions"}],"predecessor-version":[{"id":28680,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/28671\/revisions\/28680"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/media\/28672"}],"wp:attachment":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/media?parent=28671"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/categories?post=28671"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/tags?post=28671"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}