{"id":2278,"date":"2026-03-01T14:06:10","date_gmt":"2026-03-01T06:06:10","guid":{"rendered":"https:\/\/www.starverse-ai.com\/guide\/archives\/2278"},"modified":"2026-03-01T14:06:10","modified_gmt":"2026-03-01T06:06:10","slug":"ai%e7%bb%98%e7%94%bb%e5%9c%88%e7%82%b8%e5%9c%ba%ef%bc%81stable-diffusion-3-5%e7%99%bb%e9%99%86%e6%98%9f%e5%ae%87%e6%99%ba%e7%ae%97%e3%80%8c%e4%b8%80%e9%94%ae%e5%8d%b3%e7%8e%a9%e3%80%8d%ef%bc%9a","status":"publish","type":"post","link":"https:\/\/www.starverse-ai.com\/guide\/archives\/2278","title":{"rendered":"AI\u7ed8\u753b\u5708\u70b8\u573a\uff01Stable Diffusion 3.5\u767b\u9646\u661f\u5b87\u667a\u7b97\u300c\u4e00\u952e\u5373\u73a9\u300d\uff1a\u514d\u5b89\u88c5\u3001\u514d\u663e\u5361\u3001\u51fa\u56fe\u5feb3\u500d"},"content":{"rendered":"
\"AI\u7ed8\u753b\u5708\u70b8\u573a\uff01Stable<\/figure>\n

AI\u7ed8\u753b\u5708\u70b8\u573a\uff01Stable Diffusion 3.5\u767b\u9646\u661f\u5b87\u667a\u7b97\u300c\u4e00\u952e\u5373\u73a9\u300d\uff1a\u514d\u5b89\u88c5\u3001\u514d\u663e\u5361\u3001\u51fa\u56fe\u5feb3\u500d<\/h1>\n
\n

\u201c\u63d0\u793a\u8bcd\u91cc\u968f\u4fbf\u5199\u2018\u8d5b\u535a\u670b\u514b\u732b\u8033\u5973\u4ec6\u5728\u706b\u661f\u8336\u9986\u91cc\u6253\u789f\u2019\uff0cStable Diffusion 3.5 \u5c45\u7136\u4e00\u6b21\u6027\u8fd8\u539f\u4e86\u9713\u8679\u706f\u3001\u732b\u8033\u3001\u9ed1\u80f6\u673a\u548c\u7ea2\u8272\u6c99\u5c18\u66b4\u3002\u201d
\n\u8fd9\u662f\u4e0a\u5468 Reddit \u70ed\u5e16\u7684\u9ad8\u8d5e\u7559\u8a00\uff0c\u4e5f\u9053\u51fa\u4e86 SD3.5 \u53d1\u5e03\u4ec5 48 \u5c0f\u65f6\u5c31\u51b2\u4e0a GitHub Trending \u7684\u539f\u56e0\u2014\u2014\u5b98\u65b9\u5ba3\u79f0\u63d0\u793a\u8bcd\u7406\u89e3\u529b\u63d0\u5347 40%\uff0c\u591a\u4e3b\u4f53\u3001\u591a\u98ce\u683c\u3001\u591a\u5149\u7167\u7684\u590d\u6742\u573a\u666f\u4e5f\u80fd\u4e00\u6b21\u6210\u578b\u3002\u7136\u800c\u5174\u594b\u4e09\u79d2\u540e\uff0c\u5f00\u53d1\u8005\u4eec\u53c8\u96c6\u4f53\u56de\u5230\u73b0\u5b9e\uff1a\u672c\u5730 RTX 4090 24 GB \u663e\u5b58\u5f00\u6ee1\u7cbe\u5ea6\uff0c\u4f9d\u65e7\u9891\u9891 OOM\uff1b\u4e00\u5f20 1024\u00d71024 \u7684\u56fe\u8fed\u4ee3 50 \u6b65\u5c31\u8981 6 GB \u663e\u5b58\uff0c\u6279\u91cf\u8dd1 100 \u5f20\u56fe\uff0c\u663e\u5361\u98ce\u6247\u76f4\u63a5\u8d77\u98de\uff0c\u8017\u65f6 58 \u5206\u949f\uff0c\u7535\u8d39 0.4 \u5143\/\u5c0f\u65f6\uff0c\u663e\u5361\u6298\u65e7\u53e6\u7b97\u3002 <\/p>\n<\/blockquote>\n

\u6709\u6ca1\u6709\u529e\u6cd5\u201c\u65e2\u8981\u6700\u65b0\u6a21\u578b\uff0c\u53c8\u8981\u96f6\u95e8\u69db\uff0c\u8fd8\u8981\u5feb\u4e09\u500d\u201d\uff1f\u7b54\u6848\u5728\u4e91\u7aef\u2014\u2014GPU\u670d\u52a1\u5668\u79df\u7528<\/a><\/strong> \u6b63\u6210\u4e3a AI \u7ed8\u753b\u7684\u201c\u7b2c\u4e8c\u5757\u663e\u5361\u201d\u3002 <\/p>\n

\n

1. \u672c\u5730\u5361\u7687\u4e5f\u5403\u7d27\uff0c\u4e91\u7aef A100 \u624d\u662f\u201c\u5b8c\u5168\u4f53\u201d<\/h2>\n

SD3.5 \u7684\u53c2\u6570\u91cf\u4ece 1.5B \u5230 8B \u5168\u7ebf\u5347\u7ea7\uff0c\u9ed8\u8ba4\u542f\u7528 QK-Normalization\u3001Rescale CFG\uff0c\u663e\u5b58\u5360\u7528\u6c34\u6da8\u8239\u9ad8\u3002\u5b9e\u6d4b\u540c\u4e00\u53f0 i9+RTX 4090 \u5de5\u4f5c\u7ad9\uff0cPyTorch2.3+CUDA12.1 \u73af\u5883\uff0c512\u00d7512\u3001step=30\u3001batch=4\uff0c\u8fde\u7eed\u51fa 100 \u5f20\u56fe\u9700\u8981 58 \u5206\u949f\uff0c\u663e\u5b58\u5cf0\u503c 22.3 GB\uff0c\u6e29\u5ea6 83\u2103\uff0c\u98ce\u6247\u566a\u97f3 58 dB\u3002
\n\u628a\u811a\u672c\u539f\u5c01\u4e0d\u52a8\u642c\u5230 \u661f\u5b87\u667a\u7b97 GPU\u4e91\u4e3b\u673a<\/a><\/strong>\uff08A100 40GB\uff09\uff0c\u540c\u6837\u53c2\u6570\u53ea\u8981 18 \u5206\u949f\uff0c\u663e\u5b58\u5360\u7528 28 GB \u4ecd\u6709\u5bcc\u4f59\uff1b\u5355\u5f20\u56fe\u6210\u672c 0.03 \u5143\uff0c\u4e0d\u5230\u672c\u5730\u7535\u8d39\u7684\u5341\u5206\u4e4b\u4e00\u3002 <\/p>\n

\n

2. \u514d\u5b89\u88c5\u3001\u514d\u663e\u5361\uff0c\u6d4f\u89c8\u5668\u91cc\u5c31\u80fd\u201c\u5f00\u7bb1\u4f5c\u753b\u201d<\/h2>\n

\u661f\u5b87\u667a\u7b97\u628a SD3.5 + ControlNet + LoRA \u5168\u6808\u9884\u88c5\u5230\u955c\u50cf\uff0c\u7528\u6237\u53ea\u9700\uff1a
\n1. \u6ce8\u518c\u8d26\u53f7\uff08\u65b0\u7528\u6237\u9001 10 \u5143\u4f53\u9a8c\u91d1\uff09
\n2. \u9009\u62e9\u201cAI \u7ed8\u753b\u201d\u6a21\u677f\uff0c\u4e00\u952e\u542f\u52a8\u5b9e\u4f8b
\n3. \u6d4f\u89c8\u5668\u6253\u5f00 JupyterLab\uff0c\u70b9\u51fb WebUI \u94fe\u63a5\u5373\u53ef\u4f5c\u753b <\/p>\n

\u65e0\u9700\u6298\u817e Python\u3001Git\u3001CUDA \u9a71\u52a8\uff0c\u4e5f\u4e0d\u7528\u4e0b\u8f7d 20 GB \u6a21\u578b\u6743\u91cd\uff0c\u771f\u6b63\u505a\u5230\u201c\u96f6\u95e8\u69db\u201d\u3002\u5e73\u53f0\u540c\u65f6\u63d0\u4f9b\u4e91\u786c\u76d8\u3001\u4e91\u5b58\u50a8\u3001\u516c\u5171\u6a21\u578b\u5e93\u4e09\u5927\u6570\u636e\u7ba1\u7406\u7ec4\u4ef6\uff1a
\n– \u4e91\u786c\u76d8<\/a><\/strong> \u53ef\u5728\u591a\u5b9e\u4f8b\u95f4\u70ed\u63d2\u62d4\uff0c\u8bad\u7ec3\u5230\u4e00\u534a\u6362\u5361\u4e0d\u4e22\u6570\u636e
\n– \u4e91\u5b58\u50a8<\/a><\/strong> \u652f\u6301\u7f51\u9875\u76f4\u4f20\uff0c\u7d20\u6750\u968f\u7528\u968f\u53d6
\n– \u516c\u5171\u8d44\u6e90\u6c60<\/a><\/strong> \u5185\u7f6e 10 \u4e07+\u98ce\u683c\u6a21\u578b\u3001CivitAI \u70ed\u95e8 LoRA\u3001Stable Diffusion \u5b98\u65b9 Checkpoint\uff0c\u4e00\u952e\u590d\u5236\u5230\u5b9e\u4f8b\u5c31\u80fd\u8c03\u7528 <\/p>\n

\n

3. \u5b9e\u6d4b\u5bf9\u6bd4\uff1a\u672c\u5730 58 min vs \u4e91\u7aef 18 min\uff0c\u63d0\u901f 3.2 \u500d<\/h2>\n\n\n\n\n\n\n
\u73af\u5883<\/th>\nGPU<\/th>\n\u8017\u65f6<\/th>\n\u5355\u5f20\u6210\u672c<\/th>\n\u5907\u6ce8<\/th>\n<\/tr>\n<\/thead>\n
\u672c\u5730<\/td>\nRTX 4090 24 GB<\/td>\n58 min<\/td>\n0.46 \u5143*<\/td>\n\u542b\u7535\u8d39+\u6298\u65e7<\/td>\n<\/tr>\n
\u661f\u5b87\u667a\u7b97<\/td>\nA100 40 GB<\/td>\n18 min<\/td>\n0.03 \u5143<\/td>\n\u6309\u5206\u949f\u8ba1\u8d39\uff0c\u968f\u5f00\u968f\u505c<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

*\u6309 0.4 \u5143\/\u5ea6\u7535\u3001\u663e\u5361\u6298\u65e7 1.5 \u5143\/\u5929\u6298\u7b97 <\/p>\n

\u5bf9\u9700\u8981\u6279\u91cf\u51fa\u56fe\u7684\u8bbe\u8ba1\u5e08\u6765\u8bf4\uff0cGPU\u670d\u52a1\u5668\u79df\u7528<\/a><\/strong> \u7684\u5f39\u6027\u8ba1\u8d39\u76f8\u5f53\u4e8e\u201c\u628a 1.3 \u4e07\u5143\u7684 RTX 4090 \u62c6\u6210 3 \u6bdb\u94b1\u4e00\u6b21\u7684\u6309\u9700\u4ed8\u8d39\u201d\uff0c\u6210\u672c\u76f4\u964d 99%\u3002 <\/p>\n

\n

4. 10 \u4e07+\u98ce\u683c\u6a21\u578b\u76f4\u63a5\u8c03\u7528\uff0c\u7075\u611f\u4e0d\u4e2d\u65ad<\/h2>\n

\u9664\u4e86\u201c\u8dd1\u5f97\u5feb\u201d\uff0c\u661f\u5b87\u667a\u7b97\u8fd8\u60f3\u8ba9\u521b\u4f5c\u8005\u201c\u60f3\u5f97\u5c11\u201d\u3002\u5e73\u53f0\u516c\u5171\u8d44\u6e90\u5e93\u5df2\u540c\u6b65 CivitAI \u5468\u699c\u3001SD3.5 \u5b98\u65b9\u793e\u533a\u7cbe\u9009\uff0c\u6db5\u76d6\uff1a
\n– \u5199\u5b9e\u4eba\u50cf\uff1aRealVisXL\u3001JuggernautXL
\n– \u4e8c\u6b21\u5143\uff1aAnythingXL\u3001CounterfeitXL
\n– 2.5D \u6df7\u5408\uff1aDynaVisionXL
\n– \u8bbe\u8ba1\u7c7b\uff1aInteriorDesign\u3001LogoRedmond <\/p>\n

\u6240\u6709\u6a21\u578b\u4e0e LoRA \u5747\u505a\u5143\u6570\u636e\u7d22\u5f15\uff0cWebUI \u5185\u76f4\u63a5\u641c\u7d22\u5173\u952e\u8bcd\u5373\u53ef\u52a0\u8f7d\uff0c\u65e0\u9700\u624b\u52a8\u4e0b\u8f7d\u3002\u642d\u914d ControlNet \u7684 OpenPose\u3001Depth\u3001Canny \u9884\u5904\u7406\u5668\uff0c\u505a\u7535\u5546\u6a21\u7279\u3001\u52a8\u6f2b\u6d77\u62a5\u3001\u5efa\u7b51\u6e32\u67d3\u90fd\u80fd\u201c\u51fa\u56fe\u5373\u5b9a\u7a3f\u201d\u3002 <\/p>\n

\n

5. \u4e00\u5206\u949f\u4e0a\u624b\u6559\u7a0b\uff1a\u4ece\u6ce8\u518c\u5230\u51fa\u56fe\u53ea\u9700\u4e09\u6b65<\/h2>\n
    \n
  1. \u6253\u5f00 \u661f\u5b87\u667a\u7b97\u5b98\u7f51<\/a>\uff0c\u6ce8\u518c\u5373\u9886 10 \u5143\u4f53\u9a8c\u91d1 <\/li>\n
  2. \u63a7\u5236\u53f0\u9009\u62e9\u201cAI \u7ed8\u753b\u201d\u955c\u50cf\uff0cGPU \u89c4\u683c\u4efb\u9009 A100\/A6000\/RTX 4090\uff0c\u8ba1\u8d39\u6a21\u5f0f\u6309\u5206\u949f\u6216\u5305\u65e5 <\/li>\n
  3. \u542f\u52a8\u540e\u70b9\u51fb\u201cWebUI\u201d\u94fe\u63a5\uff0c\u8f93\u5165\u63d0\u793a\u8bcd\uff0c\u5373\u523b\u51fa\u56fe <\/li>\n<\/ol>\n

    \u5b9e\u4f8b\u652f\u6301\u4fdd\u5b58\u955c\u50cf\uff0c\u88c5\u597d\u81ea\u5df1\u7684 LoRA \u4e0e\u63d2\u4ef6\u540e\u53ef\u4e00\u952e\u590d\u523b\uff0c\u56e2\u961f\u534f\u4f5c\u4e5f\u80fd\u201c\u4e00\u4eba\u8c03\u901a\uff0c\u5168\u5458\u5171\u4eab\u201d\u3002 <\/p>\n

    \n

    \u7ed3\u8bed\uff1a\u628a\u663e\u5361\u9884\u7b97\u6362\u6210\u521b\u610f\u9884\u7b97<\/h2>\n

    Stable Diffusion 3.5 \u7684\u767b\u573a\u518d\u6b21\u8bc1\u660e\uff0cAI \u7ed8\u753b\u7684\u95e8\u69db\u4e0d\u5728\u6a21\u578b\uff0c\u800c\u5728\u7b97\u529b\u3002\u4e0e\u5176\u82b1\u4e0a\u4e07\u5757\u62a2\u8d2d\u663e\u5361\u3001\u71ac\u591c\u88c5\u9a71\u52a8\uff0c\u4e0d\u5982\u628a\u9884\u7b97\u6362\u6210 AI\u5e94\u7528<\/a><\/strong> \u7684\u521b\u610f\u5b9e\u9a8c\uff1a
    \n– \u5e7f\u544a\u5de5\u4f5c\u5ba4\u7528     GPU\u4e91\u4e3b\u673a<\/a><\/strong> \u4e00\u5929\u8dd1 2000 \u5f20\u7535\u5546\u56fe\uff0c\u5ba2\u6237\u73b0\u573a\u63d0\u6539\uff0c10 \u5206\u949f\u5c31\u80fd\u4ea4\u4ed8\u65b0\u56fe
    \n– \u72ec\u7acb\u5f00\u53d1\u8005\u628a\u7701\u4e0b\u7684 1 \u4e07\u5143\u786c\u4ef6\u8d39\u6295\u5165\u6295\u653e\uff0c\u4e09\u4e2a\u6708\u83b7\u5ba2\u6210\u672c\u964d 35%
    \n– \u9ad8\u6821\u8bfe\u9898\u7ec4\u6309\u9700\u79df\u7528 8\u00d7A100 \u8282\u70b9\uff0c\u8bba\u6587\u590d\u73b0\u5b8c\u5373\u91ca\u653e\uff0c\u7ecf\u8d39\u5229\u7528\u7387\u63d0\u5347 4 \u500d <\/p>\n

    \u73b0\u5728\u6ce8\u518c\u661f\u5b87\u667a\u7b97\uff0c\u65b0\u7528\u6237\u7acb\u5f97 10 \u5143\u4f53\u9a8c\u91d1\uff0c\u8db3\u591f\u514d\u8d39\u8dd1\u51fa 300 \u5f20 512\u00d7512 \u7cbe\u56fe\u3002\u4e0b\u4e00\u6b21\u7075\u611f\u7206\u53d1\u65f6\uff0c\u4f60\u53ea\u9700\u6253\u5f00\u6d4f\u89c8\u5668\uff0c\u5269\u4e0b\u7684\u4ea4\u7ed9\u4e91\u7aef\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"

    AI\u7ed8\u753b\u5708\u70b8\u573a\uff01Stable Diffusion 3.5\u767b\u9646…<\/p>\n","protected":false},"author":2,"featured_media":2277,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-2278","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-zixun"],"views":50,"_links":{"self":[{"href":"https:\/\/www.starverse-ai.com\/guide\/wp-json\/wp\/v2\/posts\/2278","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.starverse-ai.com\/guide\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.starverse-ai.com\/guide\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.starverse-ai.com\/guide\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.starverse-ai.com\/guide\/wp-json\/wp\/v2\/comments?post=2278"}],"version-history":[{"count":0,"href":"https:\/\/www.starverse-ai.com\/guide\/wp-json\/wp\/v2\/posts\/2278\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.starverse-ai.com\/guide\/wp-json\/wp\/v2\/media\/2277"}],"wp:attachment":[{"href":"https:\/\/www.starverse-ai.com\/guide\/wp-json\/wp\/v2\/media?parent=2278"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.starverse-ai.com\/guide\/wp-json\/wp\/v2\/categories?post=2278"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.starverse-ai.com\/guide\/wp-json\/wp\/v2\/tags?post=2278"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}