{"id":236,"date":"2024-01-05T11:38:57","date_gmt":"2024-01-05T03:38:57","guid":{"rendered":"https:\/\/hjwzs.xyz\/?p=236"},"modified":"2024-01-05T11:46:08","modified_gmt":"2024-01-05T03:46:08","slug":"sdxl-lora-%e8%ae%ad%e7%bb%83%e7%ac%94%e8%ae%b0","status":"publish","type":"post","link":"https:\/\/hjwzs.xyz\/?p=236","title":{"rendered":"SDXL LoRA \u8bad\u7ec3\u7b14\u8bb0"},"content":{"rendered":"\n

\u51c6\u5907\u7d20\u6750<\/h2>\n\n\n\n
    \n
  1. \u5c3a\u5bf8\uff1a\u6ca1\u6709\u8981\u6c42\u3002<\/li>\n\n\n\n
  2. \u5982\u679c\u662f\u4eba\u7269\uff0c\u9700\u8981\u591a\u89d2\u5ea6\u3002<\/li>\n<\/ol>\n\n\n\n

    \u524d\u8a00<\/h2>\n\n\n\n

    \u5982\u679c\u5728\u5927\u5c40\u57df\u7f51\u5185\uff0c\u53ef\u4ee5\u4f7f\u7528\u4ee5\u4e0b\u955c\u50cf\u4e0b\u6a21\u578b\u3001\u4ee3\u7801\u3002<\/p>\n\n\n\n

    \u4f7f\u7528\u65b9\u6cd5\u89c1\u7f51\u7ad9\u4e0a\u7684\u8bf4\u660e\u3002<\/p>\n\n\n\n

      \n
    1. https:\/\/ghproxy.com\/<\/a><\/li>\n\n\n\n
    2. https:\/\/hf-mirror.com\/<\/a><\/li>\n<\/ol>\n\n\n\n

      \u90e8\u7f72 kohya-ss<\/h2>\n\n\n\n

      \u5730\u5740\uff1ahttps:\/\/github.com\/bmaltais\/kohya_ss<\/a><\/p>\n\n\n\n

      \u5728 ubuntu \u4e0b\u90e8\u7f72<\/h3>\n\n\n\n
      $ conda create --name kohya python=3.10\n$ conda activate kohya\n# \u5b89\u88c5 cudaNN \u5728 https:\/\/developer.nvidia.com\/cuda-downloads?target_os=Linux&target_arch=x86_64\n$ git clone https:\/\/github.com\/bmaltais\/kohya_ss.git\n$ cd kohya_ss\n$ chmod +x .\/setup.sh\n$ .\/setup.sh<\/pre>\n\n\n\n
      \u542f\u52a8\u547d\u4ee4<\/h3>\n\n\n\n
      $ .\/gui.sh --listen 0.0.0.0 --server_port 5000<\/pre>\n\n\n\n
      \u7d20\u6750\u9884\u5904\u7406<\/h2>\n\n\n\n
      WebUI \u5b89\u88c5\u63d2\u4ef6\uff1ahttps:\/\/github.com\/picobyte\/stable-diffusion-webui-wd14-tagger<\/a><\/p>\n\n\n\n
        \n
      1. \u9009\u62e9 Tagger \u6807\u7b7e<\/li>\n\n\n\n
      2. \u5148\u7528\u4e00\u5f20\u56fe\u9009\u62e9\u9002\u5408\u7684\u53cd\u63a8\u7b97\u6cd5<\/li>\n\n\n\n
      3. \u8fdb\u884c\u6279\u91cf\u64cd\u4f5c\u6279\u91cf\u751f\u6210\u6253\u6807\u7684\u56fe\u7247\u96c6<\/li>\n<\/ol>\n\n\n\n
        \u6807\u7b7e\u6279\u91cf\u5904\u7406\u5de5\u5177<\/h3>\n\n\n\n
          \n
        1. windows<\/a><\/li>\n\n\n\n
        2. mac<\/a><\/li>\n<\/ol>\n\n\n\n
          \u914d\u7f6e<\/h2>\n\n\n\n
          \u6253\u5f00 http:\/\/xxx:5000<\/a><\/p>\n\n\n\n
          \u5e95\u6a21\u914d\u7f6e<\/h3>\n\n\n\n
            \n
          1. \u9009\u62e9 LoRA \u6807\u7b7e<\/li>\n\n\n\n
          2. \u5728 Traning – Source model – Model Quick Pick \u4e2d\u9009\u62e9 custom<\/li>\n\n\n\n
          3. Pretrained model name or path \u4e2d\u586b\u5165 sd_xl_base_1.0.safetensors \u7684\u8def\u5f84<\/li>\n\n\n\n
          4. \u70b9\u9009 SDXL Model<\/li>\n<\/ol>\n\n\n\n
            \u51c6\u5907\u8bad\u7ec3\u96c6<\/h3>\n\n\n\n
            \u9996\u5148\u5c06\u8bad\u7ec3\u7684\u7d20\u6750\u4e0a\u4f20\u5230\u670d\u52a1\u5668\u76ee\u5f55\u3002<\/p>\n\n\n\n
              \n
            1. \u9009\u62e9 Tools \u6807\u7b7e<\/li>\n\n\n\n
            2. \u9009\u62e9 Dataset Preparation \u6807\u7b7e<\/li>\n\n\n\n
            3. \u5728 Instance prompt \u586b\u5199 \u4e3b\u8981\u89e6\u53d1\u8bcd<\/li>\n\n\n\n
            4. \u5728 Class prompt \u586b\u5199\u5206\u7c7b\u89e6\u53d1\u8bcd<\/li>\n\n\n\n
            5. \u5728 Training images \u586b\u5199\u6570\u636e\u96c6\u8def\u5f84<\/li>\n\n\n\n
            6. \u5728 Repeats \u586b\u5199\u5faa\u73af\u6b21\u6570\uff0c\u6bd4\u5982 30 \u6b21<\/li>\n\n\n\n
            7. \u5728 Destination training directory \u586b\u5199\u8bad\u7ec3\u96c6\u8f93\u51fa\u76ee\u5f55<\/li>\n\n\n\n
            8. \u70b9\u51fb Prepare training data \u5f00\u59cb\u9884\u8bad\u7ec3<\/li>\n<\/ol>\n\n\n\n
              \u7f51\u9875\u4e0a\u4e0d\u4f1a\u6709\u63d0\u793a\uff0c\u770b\u540e\u53f0\u65e5\u5fd7\u3002\u8fd9\u4e2a\u5e94\u8be5\u5f88\u5feb\u3002<\/p>\n\n\n\n
              \u6700\u540e\u70b9\u51fb Copy info to Folders Tab \u5c06\u4fe1\u606f\u540c\u6b65\uff0c\u8fdb\u5165\u4e0b\u4e00\u6b65\u3002<\/p>\n\n\n\n
              \u8bad\u7ec3\u53c2\u6570\u8bbe\u5b9a<\/h3>\n\n\n\n
              \n
              \u8fd9\u4e2a\u8bad\u7ec3\u53c2\u6570\u9002\u5408 24G \u4ee5\u4e0a\u663e\u5b58<\/p>\n<\/blockquote>\n<\/div>\n\n\n\n
                \n
              1. \n
                \u70b9\u51fb Traning – Folders\uff0c\u68c0\u67e5\u53c2\u6570\u662f\u5426\u4ee3\u5165\u6b63\u786e<\/p>\n<\/li>\n
              2. \n
                \u586b\u5199 Model output name \u6a21\u578b\u8f93\u51fa\u540d\u5b57<\/p>\n<\/li>\n
              3. \n
                \u70b9\u51fb Parameters \u6807\u7b7e<\/p>\n<\/li>\n
              4. \n
                \u4fee\u6539\u4ee5\u4e0b\u8bbe\u7f6e\uff1a<\/p>\n
                  \n
                1. LoRA Type: LyCORIS\/LoHa<\/li>\n
                2. Train batch size: 8<\/li>\n
                3. \u52fe\u9009 Cache latents \u548c Cache latents to disk<\/li>\n
                4. Max resolution: 1024,1024<\/li>\n
                5. Epoch\uff1a5<\/li>\n
                6. Caption Extension\uff1a.txt<\/li>\n
                7. Optimizer: Prodigy<\/li>\n
                8. Learning rate: 1<\/li>\n
                9. Text Encoder learning rate: 1<\/li>\n
                10. Unet learning rate: 1<\/li>\n
                11. LR Scheduler: constant<\/li>\n
                12. LR warmup (% of steps): 0<\/li>\n
                13. Optimizer extra arguments: \u2013optimizer_args \u201csafeguard_warmup=True\u201d \u201cweight_decay=0.1\u201d \u201cdecouple=True\u201d \u201cuse_bias_correction=True\u201d \u201cd_coef=0.5\u201d –network_train_unet_only<\/strong><\/li>\n
                14. Network Rank (Dimension): 8<\/li>\n
                15. Convolution Rank (Dimension): 4<\/li>\n
                16. Network Alpha:1<\/li>\n
                17. Convolution Alpha: 1<\/li>\n
                18. Seed: 1234<\/li>\n
                19. (\u5728 Advanced \u6807\u7b7e) Noise offset:0.0357<\/li>\n
                20. (\u5728 Advanced \u6807\u7b7e) \u52fe\u9009 Gradient checkpointing<\/li>\n<\/ol>\n<\/li>\n<\/ol>\n<\/div>\n\n\n\n
                  \u6700\u540e\uff0c\u70b9\u51fb\u5f00\u59cb\u8bad\u7ec3\uff0c\u65e5\u5fd7\u5728\u540e\u53f0\uff01<\/p>\n","protected":false},"excerpt":{"rendered":"
                  \u51c6\u5907\u7d20\u6750 \u524d\u8a00 \u5982\u679c\u5728\u5927\u5c40\u57df\u7f51\u5185\uff0c\u53ef\u4ee5\u4f7f\u7528\u4ee5\u4e0b\u955c\u50cf\u4e0b\u6a21\u578b\u3001\u4ee3\u7801\u3002 \u4f7f\u7528\u65b9\u6cd5\u89c1\u7f51\u7ad9\u4e0a\u7684\u8bf4\u660e\u3002 \u90e8\u7f72 kohya- […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[5],"tags":[12,13,14],"class_list":["post-236","post","type-post","status-publish","format-standard","hentry","category-experience-sharing","tag-aigc","tag-sd","tag-sdxl"],"_links":{"self":[{"href":"https:\/\/hjwzs.xyz\/index.php?rest_route=\/wp\/v2\/posts\/236","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/hjwzs.xyz\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/hjwzs.xyz\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/hjwzs.xyz\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/hjwzs.xyz\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=236"}],"version-history":[{"count":5,"href":"https:\/\/hjwzs.xyz\/index.php?rest_route=\/wp\/v2\/posts\/236\/revisions"}],"predecessor-version":[{"id":241,"href":"https:\/\/hjwzs.xyz\/index.php?rest_route=\/wp\/v2\/posts\/236\/revisions\/241"}],"wp:attachment":[{"href":"https:\/\/hjwzs.xyz\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=236"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/hjwzs.xyz\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=236"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/hjwzs.xyz\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=236"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}