{"id":56923,"date":"2025-08-15T01:14:54","date_gmt":"2025-08-14T17:14:54","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/56923.html"},"modified":"2025-08-15T01:14:54","modified_gmt":"2025-08-14T17:14:54","slug":"qwen-image%e6%b7%b1%e5%ba%a6%e8%a7%a3%e6%9e%90%ef%bc%9a%e7%aa%81%e7%a0%b4%e6%96%87%e6%9c%ac%e6%b8%b2%e6%9f%93%e4%b8%8e%e5%9b%be%e5%83%8f%e7%bc%96%e8%be%91%e7%9a%84%e8%a7%86%e8%a7%89%e7%94%9f%e6%88%90","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/56923.html","title":{"rendered":"Qwen-Image\u6df1\u5ea6\u89e3\u6790\uff1a\u7a81\u7834\u6587\u672c\u6e32\u67d3\u4e0e\u56fe\u50cf\u7f16\u8f91\u7684\u89c6\u89c9\u751f\u6210\u5927\u6a21\u578b"},"content":{"rendered":"

Qwen-Image\u6df1\u5ea6\u89e3\u6790：\u7a81\u7834\u6587\u672c\u6e32\u67d3\u4e0e\u56fe\u50cf\u7f16\u8f91\u7684\u89c6\u89c9\u751f\u6210\u5927\u6a21\u578b<\/h3>\n

1 \u5f15\u8a00：\u89c6\u89c9\u751f\u6210\u6a21\u578b\u7684\u6311\u6218\u4e0e\u7a81\u7834<\/h4>\n

\u5728AI\u751f\u6210\u5185\u5bb9\u9886\u57df，\u56fe\u50cf\u751f\u6210\u6a21\u578b\u9762\u4e34\u4e24\u5927\u6838\u5fc3\u6311\u6218：\u590d\u6742\u6587\u672c\u6e32\u67d3\u548c\u7f16\u8f91\u4e00\u81f4\u6027\u3002\u4f20\u7edf\u6a21\u578b\u5728\u5b57\u6bcd\u6587\u5b57（\u5982\u82f1\u8bed）\u4e0a\u8868\u73b0\u5c1a\u53ef，\u4f46\u5728\u8868\u610f\u6587\u5b57（\u5982\u4e2d\u6587）\u573a\u666f\u4e2d\u5e38\u51fa\u73b0\u6587\u5b57\u626d\u66f2\u3001\u7f3a\u5931\u95ee\u9898\u3002\u540c\u65f6，\u56fe\u50cf\u7f16\u8f91\u4efb\u52a1\u4e2d\u4fdd\u6301\u539f\u59cb\u56fe\u50cf\u8bed\u4e49\u8fde\u8d2f\u6027\u548c\u89c6\u89c9\u7ec6\u8282\u4e00\u81f4\u6027\u66f4\u662f\u4e1a\u754c\u96be\u9898\u3002<\/p>\n

Qwen-Image\u4f5c\u4e3aQwen\u7cfb\u5217\u9996\u4e2a\u56fe\u50cf\u751f\u6210\u57fa\u7840\u6a21\u578b，\u901a\u8fc7\u4e09\u5927\u521b\u65b0\u7a81\u7834\u8fd9\u4e9b\u74f6\u9888：<\/p>\n

  • \u6e10\u8fdb\u5f0f\u6587\u672c\u6e32\u67d3\u8bad\u7ec3：\u4ece\u57fa\u7840\u5b57\u7b26\u5230\u6bb5\u843d\u7ea7\u63cf\u8ff0\u7684\u8bfe\u7a0b\u5b66\u4e60<\/li>\n
  • \u53cc\u7f16\u7801\u673a\u5236：Qwen2.5-VL\u8bed\u4e49\u7f16\u7801 + VAE\u89c6\u89c9\u7f16\u7801\u7684\u534f\u540c<\/li>\n
  • \u591a\u4efb\u52a1\u7edf\u4e00\u67b6\u6784：\u6587\u672c\u5230\u56fe\u50cf(T2I)\u3001\u56fe\u50cf\u5230\u56fe\u50cf(I2I)\u3001\u6587\u672c-\u56fe\u50cf\u5230\u56fe\u50cf(TI2I)\u7684\u8054\u5408\u8bad\u7ec3<\/li>\n

    \"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"<\/p>\n

    \u56fe1：Qwen-Image\u53cc\u6d41\u67b6\u6784：\u6587\u672c\u6d41（\u5de6）\u4e0e\u56fe\u50cf\u6d41（\u53f3）\u7684\u534f\u540c\u5de5\u4f5c<\/p>\n

    2 \u6a21\u578b\u67b6\u6784\u8bbe\u8ba1<\/h4>\n

    2.1 \u6574\u4f53\u6846\u67b6<\/h5>\n

    Qwen-Image\u57fa\u4e8e\u4e09\u6a21\u5757\u534f\u540c\u67b6\u6784：<\/p>\n

    class<\/span> QwenImage<\/span>(<\/span>nn.<\/span>Module)<\/span>:<\/span>
    \n def<\/span> __init__<\/span>(<\/span>self)<\/span>:<\/span>
    \n super<\/span>(<\/span>)<\/span>.<\/span>__init__(<\/span>)<\/span>
    \n self.<\/span>mllm =<\/span> Qwen2_5VL(<\/span>)<\/span> # \u591a\u6a21\u6001\u8bed\u8a00\u6a21\u578b<\/span>
    \n self.<\/span>vae =<\/span> HybridVAE(<\/span>)<\/span> # \u6df7\u5408\u53d8\u5206\u81ea\u7f16\u7801\u5668<\/span>
    \n self.<\/span>mmdit =<\/span> MMDiT(<\/span>)<\/span> # \u591a\u6a21\u6001\u6269\u6563\u53d8\u6362\u5668<\/span><\/p>\n

    def<\/span> forward<\/span>(<\/span>self,<\/span> text,<\/span> image=<\/span>None<\/span>)<\/span>:<\/span>
    \n # \u8bed\u4e49\u7279\u5f81\u63d0\u53d6<\/span>
    \n h_text =<\/span> self.<\/span>mllm(<\/span>text)<\/span> <\/p>\n

    # \u89c6\u89c9\u7279\u5f81\u63d0\u53d6<\/span>
    \n if<\/span> image is<\/span> not<\/span> None<\/span>:<\/span>
    \n z_vae =<\/span> self.<\/span>vae.<\/span>encode(<\/span>image)<\/span>
    \n h_image =<\/span> self.<\/span>mllm.<\/span>visual_encoder(<\/span>image)<\/span>
    \n h =<\/span> torch.<\/span>cat(<\/span>[<\/span>h_text,<\/span> h_image]<\/span>,<\/span> dim=<\/span>1<\/span>)<\/span>
    \n else<\/span>:<\/span>
    \n h =<\/span> h_text<\/p>\n

    # \u6269\u6563\u751f\u6210<\/span>
    \n return<\/span> self.<\/span>mmdit(<\/span>h,<\/span> z_vae)<\/span><\/p>\n

    2.2 \u591a\u6a21\u6001\u8bed\u8a00\u6a21\u578b（Qwen2.5-VL）<\/h5>\n

    \u4f5c\u4e3a\u6761\u4ef6\u7f16\u7801\u5668，\u5176\u6838\u5fc3\u4f18\u52bf\u5728\u4e8e：<\/p>\n

  • \u89c6\u89c9-\u8bed\u8a00\u7a7a\u95f4\u5bf9\u9f50<\/li>\n
  • \u4fdd\u7559\u5f3a\u8bed\u8a00\u5efa\u6a21\u80fd\u529b<\/li>\n
  • \u652f\u6301\u591a\u6a21\u6001\u8f93\u5165<\/li>\n

    \u7cfb\u7edf\u63d0\u793a\u6a21\u677f\u8bbe\u8ba1：<\/p>\n

    # \u6587\u672c\u5230\u56fe\u50cf\u4efb\u52a1\u63d0\u793a<\/span>
    \nt2i_prompt =<\/span> """
    \n<|im_start|>system
    \n\u63cf\u8ff0\u56fe\u50cf，\u8be6\u7ec6\u8bf4\u660e\u989c\u8272\u3001\u6570\u91cf\u3001\u6587\u5b57\u3001\u5f62\u72b6\u3001\u5927\u5c0f\u3001\u7eb9\u7406\u3001\u7269\u4f53\u548c\u80cc\u666f\u7684\u7a7a\u95f4\u5173\u7cfb：<|im_end|>
    \n<|im_start|>user
    \n{user_text}<|im_end|>
    \n<|im_start|>assistant
    \n"""<\/span><\/p>\n

    # \u56fe\u50cf\u7f16\u8f91\u4efb\u52a1\u63d0\u793a<\/span>
    \nt12i_prompt =<\/span> """
    \n<|im_start|>system
    \n\u63cf\u8ff0\u8f93\u5165\u56fe\u50cf\u7684\u5173\u952e\u7279\u5f81，\u7136\u540e\u8bf4\u660e\u7528\u6237\u7684\u6587\u5b57\u6307\u4ee4\u5982\u4f55\u4fee\u6539\u56fe\u50cf<|im_end|>
    \n<|im_start|>user
    \n{user_image}{user_text}<|im_end|>
    \n<|im_start|>assistant
    \n"""<\/span><\/p>\n

    2.3 \u53d8\u5206\u81ea\u7f16\u7801\u5668\u521b\u65b0<\/h5>\n

    \u91c7\u7528\u5355\u7f16\u7801\u5668\u53cc\u89e3\u7801\u5668\u67b6\u6784，\u89e3\u51b3\u4f20\u7edfVAE\u7684\u6587\u672c\u91cd\u5efa\u7f3a\u9677：<\/p>\n

    class<\/span> HybridVAE<\/span>(<\/span>nn.<\/span>Module)<\/span>:<\/span>
    \n def<\/span> __init__<\/span>(<\/span>self)<\/span>:<\/span>
    \n super<\/span>(<\/span>)<\/span>.<\/span>__init__(<\/span>)<\/span>
    \n self.<\/span>encoder =<\/span> 3DConvEncoder(<\/span>)<\/span> # \u5171\u4eab\u7f16\u7801\u5668<\/span>
    \n self.<\/span>image_decoder =<\/span> ImageDecoder(<\/span>)<\/span>
    \n self.<\/span>video_decoder =<\/span> VideoDecoder(<\/span>)<\/span><\/p>\n

    def<\/span> forward<\/span>(<\/span>self,<\/span> x,<\/span> mode=<\/span>'image'<\/span>)<\/span>:<\/span>
    \n z =<\/span> self.<\/span>encoder(<\/span>x)<\/span>
    \n if<\/span> mode ==<\/span> 'image'<\/span>:<\/span>
    \n return<\/span> self.<\/span>image_decoder(<\/span>z)<\/span>
    \n else<\/span>:<\/span> # video<\/span>
    \n return<\/span> self.<\/span>video_decoder(<\/span>z)<\/span><\/p>\n

    \u8bad\u7ec3\u5173\u952e\u53d1\u73b0：<\/p>\n

      \n
    • \u5e73\u8861\u91cd\u5efa\u635f\u5931(L1)\u4e0e\u611f\u77e5\u635f\u5931(VGG)\u51cf\u5c11\u4f2a\u5f71<\/li>\n
    • \u5bf9\u6297\u635f\u5931\u5728\u9ad8\u8d28\u91cf\u91cd\u5efa\u65f6\u6548\u679c\u51cf\u5f31<\/li>\n
    • \u4ec5\u5fae\u8c03\u89e3\u7801\u5668\u5373\u53ef\u63d0\u5347\u5c0f\u6587\u672c\u6e32\u67d3<\/li>\n<\/ul>\n
      2.4 \u591a\u6a21\u6001\u6269\u6563\u53d8\u6362\u5668<\/h5>\n

      \u91c7\u7528MMDiT\u67b6\u6784\u5e76\u521b\u65b0MSRoPE\u4f4d\u7f6e\u7f16\u7801：<\/p>\n

      class<\/span> MMDiT<\/span>(<\/span>nn.<\/span>Module)<\/span>:<\/span>
      \n def<\/span> __init__<\/span>(<\/span>self,<\/span> dim=<\/span>1024<\/span>,<\/span> heads=<\/span>16<\/span>)<\/span>:<\/span>
      \n super<\/span>(<\/span>)<\/span>.<\/span>__init__(<\/span>)<\/span>
      \n self.<\/span>layers =<\/span> nn.<\/span>ModuleList(<\/span>[<\/span>
      \n DiTBlock(<\/span>dim,<\/span> heads,<\/span> use_msrope=<\/span>True<\/span>)<\/span>
      \n for<\/span> _ in<\/span> range<\/span>(<\/span>24<\/span>)<\/span>
      \n ]<\/span>)<\/span><\/p>\n

      def<\/span> forward<\/span>(<\/span>self,<\/span> h,<\/span> z)<\/span>:<\/span>
      \n x =<\/span> self.<\/span>ms_rope(<\/span>h,<\/span> z)<\/span> # \u591a\u6a21\u6001\u4f4d\u7f6e\u7f16\u7801<\/span>
      \n for<\/span> layer in<\/span> self.<\/span>layers:<\/span>
      \n x =<\/span> layer(<\/span>x)<\/span>
      \n return<\/span> x<\/p>\n

      def<\/span> ms_rope<\/span>(<\/span>text_emb,<\/span> image_emb)<\/span>:<\/span>
      \n """
      \n \u591a\u6a21\u6001\u53ef\u6269\u5c55\u65cb\u8f6c\u4f4d\u7f6e\u7f16\u7801
      \n text_emb: [B, L_t, D]
      \n image_emb: [B, H, W, D]
      \n """<\/span>
      \n # \u6587\u672c\u89c6\u4e3a\u5bf9\u89d2\u7ebf\u5e03\u5c40<\/span>
      \n text_pos =<\/span> diag_position(<\/span>text_emb.<\/span>shape[<\/span>1<\/span>]<\/span>)<\/span>
      \n text_rot =<\/span> apply_rope(<\/span>text_emb,<\/span> text_pos)<\/span><\/p>\n

      # \u56fe\u50cf\u4e2d\u5fc3\u8d77\u59cb\u7f16\u7801<\/span>
      \n img_pos =<\/span> center_position(<\/span>image_emb.<\/span>shape[<\/span>1<\/span>:<\/span>3<\/span>]<\/span>)<\/span>
      \n img_rot =<\/span> apply_rope(<\/span>image_emb,<\/span> img_pos)<\/span><\/p>\n

      return<\/span> torch.<\/span>cat(<\/span>[<\/span>text_rot,<\/span> img_rot]<\/span>,<\/span> dim=<\/span>1<\/span>)<\/span><\/p>\n

      \u56fe2：MSRoPE\u4f4d\u7f6e\u7f16\u7801\u7b56\u7565（\u6587\u672c\u6cbf\u5bf9\u89d2\u7ebf，\u56fe\u50cf\u4ece\u4e2d\u5fc3\u8f90\u5c04）<\/p>\n

      3 \u6570\u636e\u5904\u7406\u6d41\u7a0b<\/h4>\n

      3.1 \u6570\u636e\u6536\u96c6\u7b56\u7565<\/h5>\n

      \u6784\u5efa\u767e\u4ebf\u7ea7\u56fe\u6587\u5bf9\u6570\u636e\u96c6，\u56db\u7c7b\u6570\u636e\u5206\u5e03：<\/p>\n

      #mermaid-svg-NgVuZR1w6VrznDR7 {font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;font-size:16px;fill:#333;}#mermaid-svg-NgVuZR1w6VrznDR7 .error-icon{fill:#552222;}#mermaid-svg-NgVuZR1w6VrznDR7 .error-text{fill:#552222;stroke:#552222;}#mermaid-svg-NgVuZR1w6VrznDR7 .edge-thickness-normal{stroke-width:2px;}#mermaid-svg-NgVuZR1w6VrznDR7 .edge-thickness-thick{stroke-width:3.5px;}#mermaid-svg-NgVuZR1w6VrznDR7 .edge-pattern-solid{stroke-dasharray:0;}#mermaid-svg-NgVuZR1w6VrznDR7 .edge-pattern-dashed{stroke-dasharray:3;}#mermaid-svg-NgVuZR1w6VrznDR7 .edge-pattern-dotted{stroke-dasharray:2;}#mermaid-svg-NgVuZR1w6VrznDR7 .marker{fill:#333333;stroke:#333333;}#mermaid-svg-NgVuZR1w6VrznDR7 .marker.cross{stroke:#333333;}#mermaid-svg-NgVuZR1w6VrznDR7 svg{font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;font-size:16px;}#mermaid-svg-NgVuZR1w6VrznDR7 .pieCircle{stroke:black;stroke-width:2px;opacity:0.7;}#mermaid-svg-NgVuZR1w6VrznDR7 .pieTitleText{text-anchor:middle;font-size:25px;fill:black;font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;}#mermaid-svg-NgVuZR1w6VrznDR7 .slice{font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;fill:#333;font-size:17px;}#mermaid-svg-NgVuZR1w6VrznDR7 .legend text{fill:black;font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;font-size:17px;}#mermaid-svg-NgVuZR1w6VrznDR7 :root{–mermaid-font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;}<\/p>\n

      3.2 \u4e03\u9636\u6bb5\u6570\u636e\u8fc7\u6ee4<\/h5>\n

      \u7cbe\u5bc6\u7684\u591a\u7ea7\u8fc7\u6ee4\u6d41\u7a0b：<\/p>\n

      def<\/span> multi_stage_filter<\/span>(<\/span>images,<\/span> texts,<\/span> stage)<\/span>:<\/span>
      \n if<\/span> stage ==<\/span> 1<\/span>:<\/span> # \u521d\u59cb\u6e05\u7406<\/span>
      \n images =<\/span> filter_corrupted(<\/span>images)<\/span>
      \n images =<\/span> filter_low_res(<\/span>images,<\/span> min_res=<\/span>256<\/span>)<\/span>
      \n texts =<\/span> filter_invalid_text(<\/span>texts)<\/span><\/p>\n

      elif<\/span> stage ==<\/span> 2<\/span>:<\/span> # \u8d28\u91cf\u4f18\u5316<\/span>
      \n images =<\/span> filter_blur(<\/span>images,<\/span> threshold=<\/span>0.7<\/span>)<\/span>
      \n images =<\/span> filter_low_entropy(<\/span>images,<\/span> threshold=<\/span>2.0<\/span>)<\/span><\/p>\n

      elif<\/span> stage ==<\/span> 3<\/span>:<\/span> # \u56fe\u6587\u5bf9\u9f50<\/span>
      \n pairs =<\/span> clip_alignment(<\/span>images,<\/span> texts,<\/span> threshold=<\/span>0.85<\/span>)<\/span><\/p>\n

      elif<\/span> stage ==<\/span> 4<\/span>:<\/span> # \u6587\u672c\u6e32\u67d3\u589e\u5f3a<\/span>
      \n images,<\/span> texts =<\/span> augment_text_rendering(<\/span>images,<\/span> texts)<\/span><\/p>\n

      # \u9636\u6bb55-7\u5904\u7406\u9ad8\u5206\u8fa8\u7387\u5e73\u8861<\/span>
      \n return<\/span> refined_data<\/p>\n

      3.3 \u7ed3\u6784\u5316\u6570\u636e\u6807\u6ce8<\/h5>\n

      \u521b\u65b0\u6807\u6ce8\u6846\u67b6\u540c\u6b65\u751f\u6210\u63cf\u8ff0\u4e0e\u5143\u6570\u636e：<\/p>\n

      {<\/span>
      \n "caption"<\/span>:<\/span> "\u5496\u5561\u9986\u5185\u666f，\u6728\u684c\u4e0a\u653e\u7740\u767d\u8272\u9a6c\u514b\u676f，\u676f\u8eab\u5370\u6709'Hello World'\u5b57\u6837"<\/span>,<\/span>
      \n "image_type"<\/span>:<\/span> "\u5ba4\u5185\u6444\u5f71"<\/span>,<\/span>
      \n "image_style"<\/span>:<\/span> "\u5199\u5b9e\u98ce\u683c"<\/span>,<\/span>
      \n "watermark_list"<\/span>:<\/span> [<\/span>]<\/span>,<\/span>
      \n "abnormal_elements"<\/span>:<\/span> false<\/span>
      \n}<\/span><\/p>\n

      3.4 \u6587\u672c\u6e32\u67d3\u589e\u5f3a<\/h5>\n

      \u4e09\u9636\u6bb5\u5408\u6210\u7b56\u7565\u89e3\u51b3\u4e2d\u6587\u957f\u5c3e\u5206\u5e03\u95ee\u9898：<\/p>\n

      def<\/span> render_text_synthesis<\/span>(<\/span>mode,<\/span> text,<\/span> background)<\/span>:<\/span>
      \n if<\/span> mode ==<\/span> "pure"<\/span>:<\/span> # \u7eaf\u6587\u672c\u6e32\u67d3<\/span>
      \n return<\/span> render_on_plain_bg(<\/span>text,<\/span> font_size=<\/span>24<\/span>)<\/span><\/p>\n

      elif<\/span> mode ==<\/span> "composite"<\/span>:<\/span> # \u60c5\u5883\u5408\u6210<\/span>
      \n canvas =<\/span> place_on_realistic_bg(<\/span>text,<\/span> background)<\/span>
      \n caption =<\/span> generate_context_caption(<\/span>canvas)<\/span>
      \n return<\/span> canvas,<\/span> caption<\/p>\n

      elif<\/span> mode ==<\/span> "complex"<\/span>:<\/span> # \u590d\u6742\u5e03\u5c40<\/span>
      \n template =<\/span> select_template(<\/span>"ppt"<\/span>)<\/span>
      \n return<\/span> fill_template(<\/span>template,<\/span> text)<\/span><\/p>\n

      4 \u8bad\u7ec3\u7b56\u7565\u8be6\u89e3<\/h4>\n

      4.1 \u9884\u8bad\u7ec3\u9636\u6bb5<\/h5>\n

      \u91c7\u7528\u6d41\u5339\u914d\u76ee\u6807\u51fd\u6570： <\/p>\n

      v<\/p>\n

      t<\/p>\n

      =<\/p>\n

      d<\/p>\n

      x<\/p>\n

      t<\/p>\n

      d<\/p>\n

      t<\/p>\n

      =<\/p>\n

      x<\/p>\n

      0<\/p>\n

      \u2212<\/p>\n

      x<\/p>\n

      1<\/p>\n

      v_t = \\\\frac{dx_t}{dt} = x_0 – x_1<\/p>\n

      <\/span><\/span>v<\/span><\/span>t<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>=<\/span><\/span><\/span><\/span><\/span><\/span>d<\/span>t<\/span><\/span><\/span><\/span><\/span><\/span><\/span>d<\/span>x<\/span><\/span>t<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>=<\/span><\/span><\/span><\/span>x<\/span><\/span>0<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u2212<\/span><\/span><\/span><\/span>x<\/span><\/span>1<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span> <\/p>\n

      L<\/p>\n

      =<\/p>\n

      E<\/p>\n

      \u2223<\/p>\n

      \u2223<\/p>\n

      v<\/p>\n

      \u03b8<\/p>\n

      (<\/p>\n

      x<\/p>\n

      t<\/p>\n

      ,<\/p>\n

      t<\/p>\n

      ,<\/p>\n

      h<\/p>\n

      )<\/p>\n

      \u2212<\/p>\n

      v<\/p>\n

      t<\/p>\n

      \u2223<\/p>\n

      \u2223<\/p>\n

      2<\/p>\n

      \\\\mathcal{L} = \\\\mathbb{E}||v_\\\\theta(x_t,t,h) – v_t||^2<\/p>\n

      <\/span><\/span>L<\/span><\/span>=<\/span><\/span><\/span><\/span>E<\/span>\u2223\u2223<\/span>v<\/span><\/span>\u03b8<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>(<\/span>x<\/span><\/span>t<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>,<\/span><\/span>t<\/span>,<\/span><\/span>h<\/span>)<\/span><\/span>\u2212<\/span><\/span><\/span><\/span>v<\/span><\/span>t<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u2223<\/span>\u2223<\/span><\/span>2<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n

      4.1.1 \u751f\u4ea7\u8005-\u6d88\u8d39\u8005\u6846\u67b6<\/p>\n

      # \u751f\u4ea7\u8005\u8282\u70b9<\/span>
      \ndef<\/span> producer_process<\/span>(<\/span>data_queue)<\/span>:<\/span>
      \n while<\/span> True<\/span>:<\/span>
      \n raw_data =<\/span> load_raw_data(<\/span>)<\/span>
      \n filtered =<\/span> stage_filter(<\/span>raw_data,<\/span> current_stage)<\/span>
      \n processed =<\/span> {<\/span>
      \n 'h'<\/span>:<\/span> mllm.<\/span>encode(<\/span>filtered[<\/span>'text'<\/span>]<\/span>)<\/span>,<\/span>
      \n 'z'<\/span>:<\/span> vae.<\/span>encode(<\/span>filtered[<\/span>'image'<\/span>]<\/span>)<\/span>
      \n }<\/span>
      \n data_queue.<\/span>put(<\/span>processed)<\/span><\/p>\n

      # \u6d88\u8d39\u8005\u8282\u70b9<\/span>
      \ndef<\/span> consumer_process<\/span>(<\/span>data_queue,<\/span> model)<\/span>:<\/span>
      \n while<\/span> True<\/span>:<\/span>
      \n batch =<\/span> data_queue.<\/span>get_batch(<\/span>32<\/span>)<\/span>
      \n loss =<\/span> model(<\/span>batch)<\/span>
      \n loss.<\/span>backward(<\/span>)<\/span>
      \n optimizer.<\/span>step(<\/span>)<\/span><\/p>\n

      4.1.2 \u5206\u5e03\u5f0f\u8bad\u7ec3\u4f18\u5316 \u5173\u952e\u6280\u672f\u521b\u65b0：<\/p>\n

        \n
      • 4\u8def\u5f20\u91cf\u5e76\u884c<\/li>\n
      • bfloat16\u68af\u5ea6\u805a\u5408+float32\u5f52\u7ea6<\/li>\n
      • \u7981\u7528\u6fc0\u6d3b\u68c0\u67e5\u70b9（\u8282\u770111.3%\u663e\u5b58）<\/li>\n<\/ul>\n

        4.1.3 \u6e10\u8fdb\u5f0f\u8bad\u7ec3\u7b56\u7565 \u4e94\u7ef4\u6e10\u8fdb\u5b66\u4e60：<\/p>\n

        training_strategy =<\/span> {<\/span>
        \n 'resolution'<\/span>:<\/span> [<\/span>'256×256'<\/span>,<\/span> '640×640'<\/span>,<\/span> '1328×1328'<\/span>]<\/span>,<\/span>
        \n 'text_rendering'<\/span>:<\/span> [<\/span>'none'<\/span>,<\/span> 'single'<\/span>,<\/span> 'paragraph'<\/span>]<\/span>,<\/span>
        \n 'data_quality'<\/span>:<\/span> [<\/span>'mass'<\/span>,<\/span> 'curated'<\/span>]<\/span>,<\/span>
        \n 'data_balance'<\/span>:<\/span> [<\/span>'imbalanced'<\/span>,<\/span> 'balanced'<\/span>]<\/span>,<\/span>
        \n 'data_source'<\/span>:<\/span> [<\/span>'real'<\/span>,<\/span> 'synthetic'<\/span>]<\/span>
        \n}<\/span><\/p>\n

        4.2 \u540e\u8bad\u7ec3\u4f18\u5316<\/h5>\n

        4.2.1 \u76d1\u7763\u5fae\u8c03(SFT) \u6784\u5efa\u5206\u5c42\u8bed\u4e49\u6570\u636e\u96c6：<\/p>\n

          \n
        • \u7cbe\u9009\u9ad8\u6e05\u6670\u5ea6\u3001\u7ec6\u8282\u4e30\u5bcc\u7684\u56fe\u50cf<\/li>\n
        • \u4eba\u5de5\u6807\u6ce8\u7a81\u51fa\u5173\u952e\u89c6\u89c9\u7279\u5f81<\/li>\n
        • \u5e73\u8861\u827a\u672f\u98ce\u683c\u4e0e\u771f\u5b9e\u611f\u6837\u672c<\/li>\n<\/ul>\n

          4.2.2 \u5f3a\u5316\u5b66\u4e60\u4f18\u5316 \u878d\u5408DPO\u4e0eGRPO：<\/p>\n

          # DPO\u635f\u5931\u51fd\u6570<\/span>
          \ndef<\/span> dpo_loss<\/span>(<\/span>policy_model,<\/span> ref_model,<\/span> win_data,<\/span> lose_data)<\/span>:<\/span>
          \n t =<\/span> uniform_sample(<\/span>0<\/span>,<\/span> 1<\/span>)<\/span>
          \n x_t_win =<\/span> t *<\/span> win_data.<\/span>x0 +<\/span> (<\/span>1<\/span>–<\/span>t)<\/span> *<\/span> noise
          \n v_t_win =<\/span> win_data.<\/span>x0 –<\/span> noise<\/p>\n

          diff_policy =<\/span> policy_model(<\/span>x_t_win,<\/span> t)<\/span>.<\/span>mse(<\/span>v_t_win)<\/span> –<\/span> policy_model(<\/span>x_t_lose,<\/span> t)<\/span>.<\/span>mse(<\/span>v_t_lose)<\/span>
          \n diff_ref =<\/span> ref_model(<\/span>x_t_win,<\/span> t)<\/span>.<\/span>mse(<\/span>v_t_win)<\/span> –<\/span> ref_model(<\/span>x_t_lose,<\/span> t)<\/span>.<\/span>mse(<\/span>v_t_lose)<\/span><\/p>\n

          return<\/span> –<\/span>torch.<\/span>log(<\/span>torch.<\/span>sigmoid(<\/span>–<\/span>beta*<\/span>(<\/span>diff_policy –<\/span> diff_ref)<\/span>)<\/span>)<\/span><\/p>\n

          # GRPO\u8bad\u7ec3\u6d41\u7a0b<\/span>
          \ndef<\/span> grpo_train<\/span>(<\/span>policy,<\/span> ref_model,<\/span> prompt)<\/span>:<\/span>
          \n trajectories =<\/span> [<\/span>policy.<\/span>generate(<\/span>prompt)<\/span> for<\/span> _ in<\/span> range<\/span>(<\/span>8<\/span>)<\/span>]<\/span>
          \n rewards =<\/span> reward_model(<\/span>trajectories)<\/span>
          \n advantages =<\/span> (<\/span>rewards –<\/span> rewards.<\/span>mean(<\/span>)<\/span>)<\/span> \/<\/span> rewards.<\/span>std(<\/span>)<\/span><\/p>\n

          for<\/span> traj in<\/span> trajectories:<\/span>
          \n for<\/span> t in<\/span> range<\/span>(<\/span>len<\/span>(<\/span>traj)<\/span>)<\/span>:<\/span>
          \n ratio =<\/span> policy_prob(<\/span>traj[<\/span>t]<\/span>)<\/span> \/<\/span> ref_model_prob(<\/span>traj[<\/span>t]<\/span>)<\/span>
          \n kl_div =<\/span> compute_kl(<\/span>policy,<\/span> ref_model,<\/span> traj[<\/span>t]<\/span>)<\/span>
          \n loss =<\/span> min<\/span>(<\/span>ratio*<\/span>advantages,<\/span> clip(<\/span>ratio,<\/span> 0.8<\/span>,<\/span> 1.2<\/span>)<\/span>*<\/span>advantages –<\/span> beta*<\/span>kl_div
          \n loss.<\/span>backward(<\/span>)<\/span><\/p>\n

          4.3 \u591a\u4efb\u52a1\u8bad\u7ec3\u6269\u5c55<\/h5>\n

          \u7edf\u4e00\u5904\u7406\u6846\u67b6\u652f\u6301：<\/p>\n

          def<\/span> multitask_forward<\/span>(<\/span>task_type,<\/span> text,<\/span> image=<\/span>None<\/span>)<\/span>:<\/span>
          \n if<\/span> task_type ==<\/span> 'T2I'<\/span>:<\/span>
          \n return<\/span> model.<\/span>generate_from_text(<\/span>text)<\/span><\/p>\n

          elif<\/span> task_type ==<\/span> 'TI2I'<\/span>:<\/span>
          \n h_image =<\/span> model.<\/span>encode_image(<\/span>image)<\/span>
          \n return<\/span> model.<\/span>edit_image(<\/span>text,<\/span> h_image)<\/span><\/p>\n

          elif<\/span> task_type ==<\/span> 'depth_estimation'<\/span>:<\/span>
          \n return<\/span> model.<\/span>predict_depth(<\/span>image)<\/span><\/p>\n

          elif<\/span> task_type ==<\/span> 'novel_view'<\/span>:<\/span>
          \n return<\/span> model.<\/span>render_new_view(<\/span>image,<\/span> text)<\/span><\/p>\n

          5 \u5b9e\u9a8c\u8bc4\u4f30<\/h4>\n

          5.1 \u4eba\u7c7b\u8bc4\u4f30（AI\u7ade\u6280\u573a）<\/h5>\n

          \u5efa\u7acb\u5f00\u653e\u8bc4\u4f30\u5e73\u53f0：<\/p>\n

          #mermaid-svg-PLgN2WOAD9AKTmOV {font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;font-size:16px;fill:#333;}#mermaid-svg-PLgN2WOAD9AKTmOV .error-icon{fill:#552222;}#mermaid-svg-PLgN2WOAD9AKTmOV .error-text{fill:#552222;stroke:#552222;}#mermaid-svg-PLgN2WOAD9AKTmOV .edge-thickness-normal{stroke-width:2px;}#mermaid-svg-PLgN2WOAD9AKTmOV .edge-thickness-thick{stroke-width:3.5px;}#mermaid-svg-PLgN2WOAD9AKTmOV .edge-pattern-solid{stroke-dasharray:0;}#mermaid-svg-PLgN2WOAD9AKTmOV .edge-pattern-dashed{stroke-dasharray:3;}#mermaid-svg-PLgN2WOAD9AKTmOV .edge-pattern-dotted{stroke-dasharray:2;}#mermaid-svg-PLgN2WOAD9AKTmOV .marker{fill:#333333;stroke:#333333;}#mermaid-svg-PLgN2WOAD9AKTmOV .marker.cross{stroke:#333333;}#mermaid-svg-PLgN2WOAD9AKTmOV svg{font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;font-size:16px;}#mermaid-svg-PLgN2WOAD9AKTmOV .label{font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;color:#333;}#mermaid-svg-PLgN2WOAD9AKTmOV .cluster-label text{fill:#333;}#mermaid-svg-PLgN2WOAD9AKTmOV .cluster-label span{color:#333;}#mermaid-svg-PLgN2WOAD9AKTmOV .label text,#mermaid-svg-PLgN2WOAD9AKTmOV span{fill:#333;color:#333;}#mermaid-svg-PLgN2WOAD9AKTmOV .node rect,#mermaid-svg-PLgN2WOAD9AKTmOV .node circle,#mermaid-svg-PLgN2WOAD9AKTmOV .node ellipse,#mermaid-svg-PLgN2WOAD9AKTmOV .node polygon,#mermaid-svg-PLgN2WOAD9AKTmOV .node path{fill:#ECECFF;stroke:#9370DB;stroke-width:1px;}#mermaid-svg-PLgN2WOAD9AKTmOV .node .label{text-align:center;}#mermaid-svg-PLgN2WOAD9AKTmOV .node.clickable{cursor:pointer;}#mermaid-svg-PLgN2WOAD9AKTmOV .arrowheadPath{fill:#333333;}#mermaid-svg-PLgN2WOAD9AKTmOV .edgePath .path{stroke:#333333;stroke-width:2.0px;}#mermaid-svg-PLgN2WOAD9AKTmOV .flowchart-link{stroke:#333333;fill:none;}#mermaid-svg-PLgN2WOAD9AKTmOV .edgeLabel{background-color:#e8e8e8;text-align:center;}#mermaid-svg-PLgN2WOAD9AKTmOV .edgeLabel rect{opacity:0.5;background-color:#e8e8e8;fill:#e8e8e8;}#mermaid-svg-PLgN2WOAD9AKTmOV .cluster rect{fill:#ffffde;stroke:#aaaa33;stroke-width:1px;}#mermaid-svg-PLgN2WOAD9AKTmOV .cluster text{fill:#333;}#mermaid-svg-PLgN2WOAD9AKTmOV .cluster span{color:#333;}#mermaid-svg-PLgN2WOAD9AKTmOV div.mermaidTooltip{position:absolute;text-align:center;max-width:200px;padding:2px;font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;font-size:12px;background:hsl(80, 100%, 96.2745098039%);border:1px solid #aaaa33;border-radius:2px;pointer-events:none;z-index:100;}#mermaid-svg-PLgN2WOAD9AKTmOV :root{–mermaid-font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;}<\/p>\n

          <\/span><\/p>\n

          <\/span><\/p>\n

          <\/span><\/p>\n

          <\/span><\/p>\n

          <\/span><\/p>\n

          \u751f\u6210\u63d0\u793a\u8bcd<\/p>\n

          \u6a21\u578bA\u533f\u540d\u751f\u6210<\/p>\n

          \u6a21\u578bB\u533f\u540d\u751f\u6210<\/p>\n

          \u4eba\u7c7b\u8bc4\u4f30<\/p>\n

          Elo\u8bc4\u5206\u66f4\u65b0<\/p>\n

          Qwen-Image\u5728\u4e3b\u6d41\u95ed\u6e90\u6a21\u578b\u4e2d\u6392\u540d\u7b2c\u4e8c：<\/p>\n\n\u6a21\u578bElo\u8bc4\u5206<\/tr>\n\n\n\n\n\n
          Imagen 4 Ultra<\/td>\n1250<\/td>\n<\/tr>\n
          Qwen-Image<\/td>\n1220<\/td>\n<\/tr>\n
          GPT Image 1<\/td>\n1190<\/td>\n<\/tr>\n
          FLUX.1 Kontext<\/td>\n1185<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n
          5.2 \u5b9a\u91cf\u5206\u6790<\/h5>\n

          5.2.1 VAE\u91cd\u5efa\u6027\u80fd \u5728\u6587\u672c\u91cd\u5efa\u4efb\u52a1\u4e2d\u663e\u8457\u9886\u5148：<\/p>\n

          | \u6a21\u578b | Text PSNR | SSIM |
          \n|—————–|———–|——–|
          \n| Wan2.1-VAE | 26.77 | 0.9386 |
          \n| FLUX-VAE | 32.65 | 0.9792 |
          \n| Qwen-Image-VAE | 36.63 | 0.9839 |<\/p>\n

          5.2.2 \u6587\u672c\u751f\u6210\u6027\u80fd \u4e2d\u6587\u6e32\u67d3\u78be\u538b\u7ea7\u4f18\u52bf：<\/p>\n

          | \u6a21\u578b | \u4e00\u7ea7\u6c49\u5b57 | \u4e8c\u7ea7\u6c49\u5b57 | \u4e09\u7ea7\u6c49\u5b57 |
          \n|—————|———|———|———|
          \n| Seedream 3.0 | 53.48 | 26.23 | 1.25 |
          \n| GPT Image 1 | 68.37 | 15.97 | 3.55 |
          \n| Qwen-Image | 97.29 | 40.53 | 6.48 |<\/p>\n

          5.3 \u5b9a\u6027\u5206\u6790<\/h5>\n

          5.3.1 \u4e2d\u6587\u6587\u672c\u6e32\u67d3 \"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"<\/p>\n

          \u56fe3：Qwen-Image\u7cbe\u51c6\u751f\u6210\u590d\u6742\u4e2d\u6587\u5bf9\u8054<\/p>\n

          5.3.2 \u591a\u5bf9\u8c61\u751f\u6210<\/p>\n

          \u63d0\u793a\u8bcd：\u201c\u5341\u4e8c\u751f\u8096\u6bdb\u7ed2\u73a9\u5177\u6574\u9f50\u6392\u5217，\u4e09\u884c\u56db\u5217\u201d<\/p>\n

          Qwen-Image\u552f\u4e00\u6b63\u786e\u751f\u6210\u6240\u670912\u4e2a\u751f\u8096\u5f62\u8c61\u5e76\u4fdd\u6301\u6750\u8d28\u4e00\u81f4\u6027\u3002<\/p>\n

          6 \u4ee3\u7801\u5b9e\u73b0\u5173\u952e\u6a21\u5757<\/h4>\n

          6.1 \u6d41\u5339\u914d\u8bad\u7ec3\u6838\u5fc3<\/h5>\n

          def<\/span> flow_matching_loss<\/span>(<\/span>model,<\/span> x0,<\/span> h)<\/span>:<\/span>
          \n # \u91c7\u6837\u968f\u673a\u566a\u58f0\u548c\u65f6\u95f4\u6b65<\/span>
          \n x1 =<\/span> torch.<\/span>randn_like(<\/span>x0)<\/span>
          \n t =<\/span> torch.<\/span>rand(<\/span>x0.<\/span>size(<\/span>0<\/span>)<\/span>,<\/span> device=<\/span>x0.<\/span>device)<\/span><\/p>\n

          # \u8ba1\u7b97\u4e2d\u95f4\u72b6\u6001<\/span>
          \n xt =<\/span> t[<\/span>:<\/span>,<\/span> None<\/span>,<\/span> None<\/span>,<\/span> None<\/span>]<\/span> *<\/span> x0 +<\/span> (<\/span>1<\/span>–<\/span>t)<\/span>[<\/span>:<\/span>,<\/span> None<\/span>,<\/span> None<\/span>,<\/span> None<\/span>]<\/span> *<\/span> x1
          \n vt =<\/span> x0 –<\/span> x1<\/p>\n

          # \u6a21\u578b\u9884\u6d4b<\/span>
          \n v_pred =<\/span> model(<\/span>xt,<\/span> t,<\/span> h)<\/span><\/p>\n

          return<\/span> F.<\/span>mse_loss(<\/span>v_pred,<\/span> vt)<\/span><\/p>\n

          6.2 \u751f\u4ea7\u8005-\u6d88\u8d39\u8005\u6846\u67b6<\/h5>\n

          class<\/span> DataPipeline<\/span>:<\/span>
          \n def<\/span> __init__<\/span>(<\/span>self,<\/span> num_producers=<\/span>8<\/span>)<\/span>:<\/span>
          \n self.<\/span>queue =<\/span> PriorityQueue(<\/span>maxsize=<\/span>1000<\/span>)<\/span>
          \n self.<\/span>producers =<\/span> [<\/span>Producer(<\/span>self.<\/span>queue)<\/span> for<\/span> _ in<\/span> range<\/span>(<\/span>num_producers)<\/span>]<\/span>
          \n self.<\/span>consumer =<\/span> Consumer(<\/span>self.<\/span>queue)<\/span><\/p>\n

          def<\/span> start<\/span>(<\/span>self)<\/span>:<\/span>
          \n for<\/span> p in<\/span> self.<\/span>producers:<\/span>
          \n p.<\/span>start(<\/span>)<\/span>
          \n self.<\/span>consumer.<\/span>start(<\/span>)<\/span><\/p>\n

          class<\/span> Producer<\/span>(<\/span>Thread)<\/span>:<\/span>
          \n def<\/span> run<\/span>(<\/span>self)<\/span>:<\/span>
          \n while<\/span> True<\/span>:<\/span>
          \n data =<\/span> self.<\/span>load_next_batch(<\/span>)<\/span>
          \n processed =<\/span> self.<\/span>process(<\/span>data)<\/span>
          \n self.<\/span>queue.<\/span>put(<\/span>processed,<\/span> priority=<\/span>data[<\/span>'quality_score'<\/span>]<\/span>)<\/span><\/p>\n

          class<\/span> Consumer<\/span>(<\/span>Thread)<\/span>:<\/span>
          \n def<\/span> run<\/span>(<\/span>self)<\/span>:<\/span>
          \n model =<\/span> build_model(<\/span>)<\/span>
          \n while<\/span> True<\/span>:<\/span>
          \n batch =<\/span> self.<\/span>queue.<\/span>get_batch(<\/span>32<\/span>)<\/span>
          \n loss =<\/span> model.<\/span>train_step(<\/span>batch)<\/span>
          \n update_model(<\/span>loss)<\/span><\/p>\n

          6.3 \u6e10\u8fdb\u5f0f\u8bad\u7ec3\u8c03\u5ea6\u5668<\/h5>\n

          class<\/span> ProgressiveScheduler<\/span>:<\/span>
          \n def<\/span> __init__<\/span>(<\/span>self,<\/span> total_steps)<\/span>:<\/span>
          \n self.<\/span>stage_schedule =<\/span> {<\/span>
          \n 'resolution'<\/span>:<\/span> [<\/span>
          \n (<\/span>0<\/span>,<\/span> 0.3<\/span>,<\/span> 256<\/span>)<\/span>,<\/span>
          \n (<\/span>0.3<\/span>,<\/span> 0.7<\/span>,<\/span> 640<\/span>)<\/span>,<\/span>
          \n (<\/span>0.7<\/span>,<\/span> 1.0<\/span>,<\/span> 1328<\/span>)<\/span>
          \n ]<\/span>,<\/span>
          \n 'text_complexity'<\/span>:<\/span> [<\/span>
          \n (<\/span>0<\/span>,<\/span> 0.4<\/span>,<\/span> 'word'<\/span>)<\/span>,<\/span>
          \n (<\/span>0.4<\/span>,<\/span> 0.8<\/span>,<\/span> 'line'<\/span>)<\/span>,<\/span>
          \n (<\/span>0.8<\/span>,<\/span> 1.0<\/span>,<\/span> 'paragraph'<\/span>)<\/span>
          \n ]<\/span>
          \n }<\/span><\/p>\n

          def<\/span> update<\/span>(<\/span>self,<\/span> step,<\/span> total)<\/span>:<\/span>
          \n progress =<\/span> step \/<\/span> total
          \n config =<\/span> {<\/span>}<\/span><\/p>\n

          for<\/span> key,<\/span> stages in<\/span> self.<\/span>stage_schedule.<\/span>items(<\/span>)<\/span>:<\/span>
          \n for<\/span> start,<\/span> end,<\/span> value in<\/span> stages:<\/span>
          \n if<\/span> start <=<\/span> progress <<\/span> end:<\/span>
          \n config[<\/span>key]<\/span> =<\/span> value
          \n break<\/span><\/p>\n

          return<\/span> config<\/p>\n

          7 \u5e94\u7528\u573a\u666f\u4e0e\u672a\u6765\u65b9\u5411<\/h4>\n

          7.1 \u4ea7\u4e1a\u5e94\u7528\u4ef7\u503c<\/h5>\n
        • \u8bbe\u8ba1\u9886\u57df：\u81ea\u52a8\u751f\u6210\u6d77\u62a5\u3001UI\u754c\u9762generate_poster(<\/span>title=<\/span>"\u590f\u5b63\u4fc3\u9500"<\/span>,<\/span>
          \n elements=<\/span>[<\/span>"\u6298\u6263\u6807\u7b7e"<\/span>,<\/span> "\u4ea7\u54c1\u56fe\u7247"<\/span>,<\/span> "\u8054\u7cfb\u65b9\u5f0f"<\/span>]<\/span>,<\/span>
          \n style=<\/span>"\u73b0\u4ee3\u7b80\u7ea6"<\/span>)<\/span>\n <\/li>\n
        • \u7535\u5b50\u5546\u52a1：\u591a\u8bed\u8a00\u5546\u54c1\u56fe\u6587\u751f\u6210<\/li>\n
        • \u6559\u80b2\u5185\u5bb9：\u56fe\u6587\u5e76\u8302\u7684\u6559\u5b66\u6750\u6599<\/li>\n
          7.2 \u6280\u672f\u6f14\u8fdb\u65b9\u5411<\/h5>\n
        • \u89c6\u9891\u751f\u6210\u6269\u5c55：\u5229\u7528VAE\u89c6\u9891\u89e3\u7801\u5668<\/li>\n
        • 3D\u751f\u6210\u7edf\u4e00：\u65b0\u89c6\u89d2\u5408\u6210\u6280\u672f\u5ef6\u4f38generate_3d_asset(<\/span>image,<\/span> prompt=<\/span>"\u751f\u6210360\u5ea6\u65cb\u8f6c\u89c6\u56fe"<\/span>)<\/span>\n <\/li>\n
        • \u611f\u77e5-\u751f\u6210\u4e00\u4f53\u5316：\u4e0eQwen-VL\u89c6\u89c9\u7406\u89e3\u6a21\u578b\u878d\u5408<\/li>\n

          \u521b\u65b0\u542f\u793a：Qwen-Image\u7684\u6280\u672f\u7a81\u7834\u8bc1\u660e，\u751f\u6210\u6a21\u578b\u53ef\u901a\u8fc7\u5bf9\u5e95\u5c42\u5206\u5e03\u7684\u5efa\u6a21\u5b9e\u73b0\u6df1\u5ea6\u7406\u89e3，\u8fd9\u6a21\u7cca\u4e86\u751f\u6210\u4e0e\u7406\u89e3\u7684\u754c\u9650，\u4e3a\u591a\u6a21\u6001AGI\u5f00\u8f9f\u4e86\u65b0\u8def\u5f84\u3002<\/p>\n

          \u7ed3\u8bba<\/h4>\n

          Qwen-Image\u901a\u8fc7\u4e09\u9879\u6839\u672c\u6027\u521b\u65b0\u91cd\u5851\u56fe\u50cf\u751f\u6210\u8303\u5f0f：<\/p>\n

        • \u6587\u672c\u6e32\u67d3\u9769\u65b0：\u6e10\u8fdb\u5f0f\u8bad\u7ec3\u7b56\u7565\u653b\u514b\u8868\u610f\u6587\u5b57\u96be\u9898<\/li>\n
        • \u7f16\u8f91\u4e00\u81f4\u6027\u7a81\u7834：\u53cc\u7f16\u7801\u673a\u5236\u5e73\u8861\u8bed\u4e49\u4e0e\u89c6\u89c9<\/li>\n
        • \u7edf\u4e00\u67b6\u6784\u8bbe\u8ba1：MMDiT\u9aa8\u5e72\u7f51\u7edc\u652f\u6301\u591a\u4efb\u52a1\u751f\u6210<\/li>\n

          \u5176\u6280\u672f\u5f71\u54cd\u5df2\u8d85\u8d8a\u56fe\u50cf\u751f\u6210\u9886\u57df，\u9884\u793a\u4e86\u4e09\u4e2a\u672a\u6765\u65b9\u5411：<\/p>\n

        • \u751f\u6210\u5373\u7406\u89e3：\u901a\u8fc7\u751f\u6210\u8fc7\u7a0b\u5b9e\u73b0\u6df1\u5ea6\u89c6\u89c9\u7406\u89e3<\/li>\n
        • \u591a\u6a21\u6001\u7edf\u4e00：\u6587\u672c\u3001\u56fe\u50cf\u3001\u89c6\u9891\u7684\u8054\u5408\u5efa\u6a21<\/li>\n
        • \u4ea4\u4e92\u8303\u5f0f\u53d8\u9769：\u89c6\u89c9\u8bed\u8a00\u754c\u9762(VLI)\u53d6\u4ee3\u4f20\u7edfGUI<\/li>\n

          \u968f\u7740\u4ee3\u7801\u5f00\u6e90（GitHub\u94fe\u63a5\u548c\u6a21\u578b\u6743\u91cd\u91ca\u653e，Qwen-Image\u5c06\u6210\u4e3a\u751f\u6210\u5f0fAI\u53d1\u5c55\u7684\u91cd\u8981\u91cc\u7a0b\u7891\u3002<\/p>\n

          \n

          \u53c2\u8003\u6587\u732e：<\/p>\n

        • Qwen-Image Technical Report<\/li>\n
        • Esser P, et al. \u201cScaling Rectified Flow Transformers for High-Resolution Image Synthesis\u201d ICML 2024<\/li>\n
        • HuggingFace Diffusers Library<\/li>\n
        • MMDiT: Multi-Modal Diffusion Transformer<\/li>\n","protected":false},"excerpt":{"rendered":"

          \u6587\u7ae0\u6d4f\u89c8\u9605\u8bfb7k\u6b21\uff0c\u70b9\u8d5e26\u6b21\uff0c\u6536\u85cf20\u6b21\u3002Qwen-Image\u662f\u4e00\u6b3e\u7a81\u7834\u6027\u7684\u89c6\u89c9\u751f\u6210\u6a21\u578b\uff0c\u9488\u5bf9\u6587\u672c\u6e32\u67d3\u548c\u56fe\u50cf\u7f16\u8f91\u4e24\u5927\u6838\u5fc3\u6311\u6218\u63d0\u51fa\u521b\u65b0\u89e3\u51b3\u65b9\u6848\u3002\u8be5\u6a21\u578b\u91c7\u7528\u6e10\u8fdb\u5f0f\u6587\u672c\u6e32\u67d3\u8bad\u7ec3\u548c\u53cc\u7f16\u7801\u673a\u5236\uff08Qwen2.5-VL\u8bed\u4e49\u7f16\u7801+VAE\u89c6\u89c9\u7f16\u7801\uff09\uff0c\u901a\u8fc7\u591a\u4efb\u52a1\u7edf\u4e00\u67b6\u6784\u5b9e\u73b0\u6587\u672c\u5230\u56fe\u50cf\u3001\u56fe\u50cf\u5230\u56fe\u50cf\u7b49\u529f\u80fd\u7684\u8054\u5408\u8bad\u7ec3\u3002\u5176\u6838\u5fc3\u521b\u65b0\u5305\u62ec\uff1a\u591a\u6a21\u6001\u8bed\u8a00\u6a21\u578b\u4f5c\u4e3a\u6761\u4ef6\u7f16\u7801\u5668\uff0c\u6df7\u5408\u53d8\u5206\u81ea\u7f16\u7801\u5668\u89e3\u51b3\u6587\u672c\u91cd\u5efa\u7f3a\u9677\uff0c\u4ee5\u53ca\u91c7\u7528MMDiT\u67b6\u6784\u548c\u521b\u65b0\u7684MSRoPE\u4f4d\u7f6e\u7f16\u7801\u3002\u6a21\u578b\u57fa\u4e8e\u767e\u4ebf\u7ea7\u56fe\u6587\u5bf9\u6570\u636e\u96c6\uff0c\u7ecf\u8fc7\u4e03\u9636\u6bb5\u7cbe\u5bc6\u8fc7\u6ee4\u6d41\u7a0b\uff0c\u7279\u522b\u9488\u5bf9\u4e2d\u6587\u7b49\u8868\u610f\u6587\u5b57\u7684\u6e32\u67d3\u95ee\u9898\u8bbe\u8ba1\u4e86\u6587\u672c\u5408\u6210\u589e\u5f3a\u7b56\u7565\u3002\u8bad\u7ec3\u91c7\u7528\u6d41\u5339\u914d\u76ee\u6807\u51fd<\/p>\n","protected":false},"author":2,"featured_media":56921,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[52,5929,50,227,4378],"topic":[],"class_list":["post-56923","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-server","tag-aigc","tag-qwen-image","tag-50","tag-227","tag-4378"],"yoast_head":"\nQwen-Image\u6df1\u5ea6\u89e3\u6790\uff1a\u7a81\u7834\u6587\u672c\u6e32\u67d3\u4e0e\u56fe\u50cf\u7f16\u8f91\u7684\u89c6\u89c9\u751f\u6210\u5927\u6a21\u578b - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wsisp.com\/helps\/56923.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Qwen-Image\u6df1\u5ea6\u89e3\u6790\uff1a\u7a81\u7834\u6587\u672c\u6e32\u67d3\u4e0e\u56fe\u50cf\u7f16\u8f91\u7684\u89c6\u89c9\u751f\u6210\u5927\u6a21\u578b - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\u6587\u7ae0\u6d4f\u89c8\u9605\u8bfb7k\u6b21\uff0c\u70b9\u8d5e26\u6b21\uff0c\u6536\u85cf20\u6b21\u3002Qwen-Image\u662f\u4e00\u6b3e\u7a81\u7834\u6027\u7684\u89c6\u89c9\u751f\u6210\u6a21\u578b\uff0c\u9488\u5bf9\u6587\u672c\u6e32\u67d3\u548c\u56fe\u50cf\u7f16\u8f91\u4e24\u5927\u6838\u5fc3\u6311\u6218\u63d0\u51fa\u521b\u65b0\u89e3\u51b3\u65b9\u6848\u3002\u8be5\u6a21\u578b\u91c7\u7528\u6e10\u8fdb\u5f0f\u6587\u672c\u6e32\u67d3\u8bad\u7ec3\u548c\u53cc\u7f16\u7801\u673a\u5236\uff08Qwen2.5-VL\u8bed\u4e49\u7f16\u7801+VAE\u89c6\u89c9\u7f16\u7801\uff09\uff0c\u901a\u8fc7\u591a\u4efb\u52a1\u7edf\u4e00\u67b6\u6784\u5b9e\u73b0\u6587\u672c\u5230\u56fe\u50cf\u3001\u56fe\u50cf\u5230\u56fe\u50cf\u7b49\u529f\u80fd\u7684\u8054\u5408\u8bad\u7ec3\u3002\u5176\u6838\u5fc3\u521b\u65b0\u5305\u62ec\uff1a\u591a\u6a21\u6001\u8bed\u8a00\u6a21\u578b\u4f5c\u4e3a\u6761\u4ef6\u7f16\u7801\u5668\uff0c\u6df7\u5408\u53d8\u5206\u81ea\u7f16\u7801\u5668\u89e3\u51b3\u6587\u672c\u91cd\u5efa\u7f3a\u9677\uff0c\u4ee5\u53ca\u91c7\u7528MMDiT\u67b6\u6784\u548c\u521b\u65b0\u7684MSRoPE\u4f4d\u7f6e\u7f16\u7801\u3002\u6a21\u578b\u57fa\u4e8e\u767e\u4ebf\u7ea7\u56fe\u6587\u5bf9\u6570\u636e\u96c6\uff0c\u7ecf\u8fc7\u4e03\u9636\u6bb5\u7cbe\u5bc6\u8fc7\u6ee4\u6d41\u7a0b\uff0c\u7279\u522b\u9488\u5bf9\u4e2d\u6587\u7b49\u8868\u610f\u6587\u5b57\u7684\u6e32\u67d3\u95ee\u9898\u8bbe\u8ba1\u4e86\u6587\u672c\u5408\u6210\u589e\u5f3a\u7b56\u7565\u3002\u8bad\u7ec3\u91c7\u7528\u6d41\u5339\u914d\u76ee\u6807\u51fd\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/56923.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2025-08-14T17:14:54+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/08\/20250814171450-689e198ae2a8a.png\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"8 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/56923.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/56923.html\",\"name\":\"Qwen-Image\u6df1\u5ea6\u89e3\u6790\uff1a\u7a81\u7834\u6587\u672c\u6e32\u67d3\u4e0e\u56fe\u50cf\u7f16\u8f91\u7684\u89c6\u89c9\u751f\u6210\u5927\u6a21\u578b - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2025-08-14T17:14:54+00:00\",\"dateModified\":\"2025-08-14T17:14:54+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/56923.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/56923.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/56923.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Qwen-Image\u6df1\u5ea6\u89e3\u6790\uff1a\u7a81\u7834\u6587\u672c\u6e32\u67d3\u4e0e\u56fe\u50cf\u7f16\u8f91\u7684\u89c6\u89c9\u751f\u6210\u5927\u6a21\u578b\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Qwen-Image\u6df1\u5ea6\u89e3\u6790\uff1a\u7a81\u7834\u6587\u672c\u6e32\u67d3\u4e0e\u56fe\u50cf\u7f16\u8f91\u7684\u89c6\u89c9\u751f\u6210\u5927\u6a21\u578b - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.wsisp.com\/helps\/56923.html","og_locale":"zh_CN","og_type":"article","og_title":"Qwen-Image\u6df1\u5ea6\u89e3\u6790\uff1a\u7a81\u7834\u6587\u672c\u6e32\u67d3\u4e0e\u56fe\u50cf\u7f16\u8f91\u7684\u89c6\u89c9\u751f\u6210\u5927\u6a21\u578b - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\u6587\u7ae0\u6d4f\u89c8\u9605\u8bfb7k\u6b21\uff0c\u70b9\u8d5e26\u6b21\uff0c\u6536\u85cf20\u6b21\u3002Qwen-Image\u662f\u4e00\u6b3e\u7a81\u7834\u6027\u7684\u89c6\u89c9\u751f\u6210\u6a21\u578b\uff0c\u9488\u5bf9\u6587\u672c\u6e32\u67d3\u548c\u56fe\u50cf\u7f16\u8f91\u4e24\u5927\u6838\u5fc3\u6311\u6218\u63d0\u51fa\u521b\u65b0\u89e3\u51b3\u65b9\u6848\u3002\u8be5\u6a21\u578b\u91c7\u7528\u6e10\u8fdb\u5f0f\u6587\u672c\u6e32\u67d3\u8bad\u7ec3\u548c\u53cc\u7f16\u7801\u673a\u5236\uff08Qwen2.5-VL\u8bed\u4e49\u7f16\u7801+VAE\u89c6\u89c9\u7f16\u7801\uff09\uff0c\u901a\u8fc7\u591a\u4efb\u52a1\u7edf\u4e00\u67b6\u6784\u5b9e\u73b0\u6587\u672c\u5230\u56fe\u50cf\u3001\u56fe\u50cf\u5230\u56fe\u50cf\u7b49\u529f\u80fd\u7684\u8054\u5408\u8bad\u7ec3\u3002\u5176\u6838\u5fc3\u521b\u65b0\u5305\u62ec\uff1a\u591a\u6a21\u6001\u8bed\u8a00\u6a21\u578b\u4f5c\u4e3a\u6761\u4ef6\u7f16\u7801\u5668\uff0c\u6df7\u5408\u53d8\u5206\u81ea\u7f16\u7801\u5668\u89e3\u51b3\u6587\u672c\u91cd\u5efa\u7f3a\u9677\uff0c\u4ee5\u53ca\u91c7\u7528MMDiT\u67b6\u6784\u548c\u521b\u65b0\u7684MSRoPE\u4f4d\u7f6e\u7f16\u7801\u3002\u6a21\u578b\u57fa\u4e8e\u767e\u4ebf\u7ea7\u56fe\u6587\u5bf9\u6570\u636e\u96c6\uff0c\u7ecf\u8fc7\u4e03\u9636\u6bb5\u7cbe\u5bc6\u8fc7\u6ee4\u6d41\u7a0b\uff0c\u7279\u522b\u9488\u5bf9\u4e2d\u6587\u7b49\u8868\u610f\u6587\u5b57\u7684\u6e32\u67d3\u95ee\u9898\u8bbe\u8ba1\u4e86\u6587\u672c\u5408\u6210\u589e\u5f3a\u7b56\u7565\u3002\u8bad\u7ec3\u91c7\u7528\u6d41\u5339\u914d\u76ee\u6807\u51fd","og_url":"https:\/\/www.wsisp.com\/helps\/56923.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2025-08-14T17:14:54+00:00","og_image":[{"url":"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/08\/20250814171450-689e198ae2a8a.png"}],"author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"8 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/56923.html","url":"https:\/\/www.wsisp.com\/helps\/56923.html","name":"Qwen-Image\u6df1\u5ea6\u89e3\u6790\uff1a\u7a81\u7834\u6587\u672c\u6e32\u67d3\u4e0e\u56fe\u50cf\u7f16\u8f91\u7684\u89c6\u89c9\u751f\u6210\u5927\u6a21\u578b - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2025-08-14T17:14:54+00:00","dateModified":"2025-08-14T17:14:54+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/56923.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/56923.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/56923.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"Qwen-Image\u6df1\u5ea6\u89e3\u6790\uff1a\u7a81\u7834\u6587\u672c\u6e32\u67d3\u4e0e\u56fe\u50cf\u7f16\u8f91\u7684\u89c6\u89c9\u751f\u6210\u5927\u6a21\u578b"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/56923","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=56923"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/56923\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media\/56921"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=56923"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=56923"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=56923"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=56923"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}