{"id":3159,"date":"2026-03-11T14:06:30","date_gmt":"2026-03-11T06:06:30","guid":{"rendered":"https:\/\/www.starverse-ai.com\/guide\/archives\/3159"},"modified":"2026-03-11T14:06:30","modified_gmt":"2026-03-11T06:06:30","slug":"stable-diffusion-3-%e5%bc%80%e6%ba%90%e5%bc%95%e7%88%86%e8%ae%be%e8%ae%a1%e5%b8%88%e5%9c%88%ef%bc%8c%e6%98%9f%e5%ae%87%e6%99%ba%e7%ae%97ai%e7%bb%98%e7%94%bb%e4%b8%80%e9%94%ae%e5%8d%b3","status":"publish","type":"post","link":"https:\/\/www.starverse-ai.com\/guide\/archives\/3159","title":{"rendered":"Stable Diffusion 3 \u5f00\u6e90\u5f15\u7206\u8bbe\u8ba1\u5e08\u5708\uff0c\u661f\u5b87\u667a\u7b97\u201cAI\u7ed8\u753b\u4e00\u952e\u5373\u73a9\u201d\u5b9e\u6d4b\uff1aRTX 4090 \u663e\u5361\u79df\u7528 1 \u5143\u8d77\uff0c\u51fa\u56fe\u901f\u5ea6\u7ffb\u500d"},"content":{"rendered":"
\"Stable<\/figure>\n
\n

\u201c\u51cc\u66682\u70b9\uff0cStable Diffusion 3 \u6743\u91cd\u521a\u653e\uff0cDiscord \u7fa4\u91cc 500 \u4f4d\u8bbe\u8ba1\u5e08\u540c\u65f6\u558a\uff1a\u201824G \u663e\u5b58\u70b8\u4e86\uff01\u2019\u201d
\n\u2014\u2014 \u4e00\u6761\u51cc\u6668\u7684 Reddit \u70ed\u5e16\uff0c48 \u5c0f\u65f6\u6536\u83b7 12K \u70b9\u8d5e\u3002<\/p>\n<\/blockquote>\n

\u70ed\u70b9\uff1aSD3 \u6a21\u578b\u6743\u91cd\u521a\u653e\uff0c\u8bbe\u8ba1\u5e08\u8fde\u591c\u6392\u961f\u4e0b\u8f7d<\/h2>\n

6 \u6708 12 \u65e5\uff0cStability AI \u5728 Hugging Face \u6084\u6084\u4e0a\u4f20\u4e86 SD3-4.7B \u7684 safetensors\u3002\u4e0d\u5230 30 \u5206\u949f\uff0c\u4e0b\u8f7d\u699c\u5355\u5c31\u88ab\u201cChinese_Designer_01\u201d\u201cFreelance_99\u201d\u7b49 ID \u5237\u5c4f\u3002\u539f\u56e0\u65e0\u4ed6\uff1aSD3 \u5728 4K \u91c7\u6837\u4e0b\u652f\u6301\u591a\u4e3b\u4f53\u3001\u5d4c\u5165\u4e2d\u82f1\u53cc\u8bed\u6392\u7248\uff0c\u5546\u4e1a\u6d77\u62a5\u53ef\u76f4\u63a5\u51fa\u56fe\u3002\u4f46\u72c2\u6b22\u53ea\u6301\u7eed\u4e86 10 \u5206\u949f\u2014\u2014\u9996\u6279\u5c1d\u9c9c\u8005\u628a 16 \u5f20 4096\u00d74096 \u56fe\u585e\u8fdb Pipeline\uff0c\u672c\u5730 RTX 3080 24G \u77ac\u95f4 OOM\uff0c\u9ed1\u5c4f\u91cd\u542f\uff0cPS \u672a\u4fdd\u5b58\u7684 3 \u5c0f\u65f6\u8349\u56fe\u76f4\u63a5\u84b8\u53d1\u3002\u6709\u4eba\u5410\u69fd\uff1a\u201c\u8fd9\u4e0d\u662f AI \u7ed8\u753b\uff0c\u662f\u663e\u5361\u706b\u5316\u3002\u201d<\/p>\n

\u75db\u70b9\uff1a\u672c\u5730 24G \u663e\u5b58\u544a\u6025\uff0c\u51fa 4K \u56fe\u76f4\u63a5\u9ed1\u5c4f<\/h2>\n

\u6211\u4eec\u628a\u955c\u5934\u62c9\u56de\u56fd\u5185\u3002\u4e0a\u6d77\u72ec\u7acb\u54c1\u724c\u8bbe\u8ba1\u5e08\u963f\u5353\u7b97\u4e86\u4e00\u7b14\u8d26\uff1a
\n– \u672c\u5730 RTX 3080 24G\uff0c\u5355\u5f20 4K \u56fe 150 \u6b65\u91c7\u6837 \u2248 8 \u5206\u949f
\n– \u4e00\u6b21 Campaign \u9700\u6c42 100 \u5f20\uff0c\u5c31\u8981 800 \u5206\u949f \u2248 13.3 \u5c0f\u65f6
\n– \u6ee1\u8f7d\u529f\u8017 320W\uff0c\u7535\u8d39 0.65 \u5143\/\u5ea6\uff0c\u5355\u4efb\u52a1 2.8 \u5143
\n– \u663e\u5361\u6bcf\u5929 13 \u5c0f\u65f6 100% \u8d1f\u8f7d\uff0c3 \u4e2a\u6708\u6298\u65e7\u8dcc\u4ef7 30%\uff0c\u6298\u5408 2.7 \u5143\/\u5c0f\u65f6 <\/p>\n

\u66f4\u9690\u6027\u7684\u662f\u201c\u7b49\u5f85\u6210\u672c\u201d\u2014\u2014\u5ba2\u6237\u5728\u5fae\u4fe1\u90a3\u5934\u4e0d\u505c\u5237\u65b0\u201c\u597d\u4e86\u5417\uff1f\u201d\u6bcf\u5ef6\u8fdf 1 \u5c0f\u65f6\uff0c\u5c3e\u6b3e\u98ce\u9669\u63d0\u9ad8 5%\u3002\u663e\u5b58\u544a\u6025\u3001\u98ce\u6247\u72c2\u8f6c\u3001\u7535\u6e90\u7126\u5473\uff0c\u6210\u4e3a\u81ea\u7531\u8bbe\u8ba1\u5e08\u7684\u201c\u4e0d\u53ef\u80fd\u4e09\u89d2\u201d\u3002<\/p>\n

\u65b9\u6848\uff1aGPU\u670d\u52a1\u5668\u79df\u7528 4090 \u8282\u70b9\uff0c\u661f\u5b87\u667a\u7b97\u5e73\u53f0\u9884\u88c5 SD3 WebUI\uff0c\u70b9\u51fb\u5373\u73a9<\/h2>\n

\u963f\u5353\u5728\u670b\u53cb\u5708\u5237\u5230\u201cGPU\u4e91\u4e3b\u673a<\/a> 1 \u5143\u8d77\u79df\u201d\u7684\u6d77\u62a5\uff0c\u70b9\u8fdb\u661f\u5b87\u667a\u7b97<\/a>\u63a7\u5236\u53f0\uff0c\u53d1\u73b0\u4ed6\u4eec\u5df2\u7ecf\u628a SD3 \u73af\u5883\u505a\u6210\u4e86\u201c\u5e94\u7528\u6a21\u677f\u201d\uff1a
\n1. \u9009\u62e9\u300cRTX 4090 24G\u300d\u5b9e\u4f8b\uff0c\u7cfb\u7edf\u81ea\u52a8\u6302\u8f7d CUDA 12.2\u3001PyTorch 2.3\u3001xFormers 0.0.22
\n2. 30 \u79d2\u5f00\u673a\u5b8c\u6210\uff0cWebUI \u5730\u5740+Token \u4e00\u952e\u590d\u5236\u5230\u6d4f\u89c8\u5668
\n3. \u6a21\u578b\u5e93\u91cc\u52fe\u9009 sd3_4.7b.safetensors\uff0c\u70b9\u51fb\u300c\u542f\u52a8\u300d\u5373\u53ef\u51fa\u56fe <\/p>\n

\u6574\u4e2a\u6d41\u7a0b 3 \u5206\u949f\uff0c\u6bd4\u672c\u5730\u89e3\u538b\u6743\u91cd\u8fd8\u5feb\u3002\u963f\u5353\u611f\u53f9\uff1a\u201c\u539f\u6765GPU\u670d\u52a1\u5668\u79df\u7528<\/a>\u4e5f\u80fd\u50cf\u70b9\u5916\u5356\u4e00\u6837\u7b80\u5355\u3002\u201d<\/p>\n

\u6570\u636e\uff1a\u5bf9\u6bd4\u672c\u5730 3080 \u4e0e\u4e91\u7aef 4090\uff0c\u6bcf 100 \u5f20 4K \u56fe\u65f6\u95f4\/\u7535\u8d39\/\u786c\u4ef6\u6298\u65e7\u6210\u672c<\/h2>\n

\u6211\u4eec\u628a\u963f\u5353\u7684\u5b9e\u6d4b\u6574\u7406\u6210\u8868\uff0c\u91c7\u6837\u53c2\u6570\u7edf\u4e00\uff1a4K \u5206\u8fa8\u7387\u3001150 steps\u3001DPM++ 2M Karras\u3001CFG 7.5\u3002<\/p>\n\n\n\n\n\n\n
\u573a\u666f<\/th>\n\u65f6\u95f4<\/th>\n\u7535\u8d39<\/th>\n\u786c\u4ef6\u6298\u65e7<\/th>\n\u603b\u6210\u672c<\/th>\n\u5e73\u5747\u5355\u5f20<\/th>\n<\/tr>\n<\/thead>\n
\u672c\u5730 RTX 3080 24G<\/td>\n13.3 h<\/td>\n2.8 \u5143<\/td>\n36.0 \u5143<\/td>\n38.8 \u5143<\/td>\n0.388 \u5143<\/td>\n<\/tr>\n
\u661f\u5b87\u667a\u7b97 RTX 4090<\/td>\n5.8 h<\/td>\n1.3 \u5143<\/td>\n0 \u5143<\/td>\n6.2 \u5143<\/td>\n0.062 \u5143<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n