{"id":3370,"date":"2023-12-06T17:37:21","date_gmt":"2023-12-06T09:37:21","guid":{"rendered":"http:\/\/xinyixx.com\/?p=3370"},"modified":"2023-12-10T18:50:22","modified_gmt":"2023-12-10T10:50:22","slug":"sdxl-turbo-2","status":"publish","type":"post","link":"https:\/\/www.xinyixx.com\/index.php\/2023\/12\/06\/sdxl-turbo-2\/","title":{"rendered":"ComfyUI\u79d2\u51fa\u56fe-SDXL-Turbo"},"content":{"rendered":"<p>\u539f\u6587\uff1a<a href=\"https:\/\/zhuanlan.zhihu.com\/p\/670221383\" target=\"_blank\" rel=\"noopener\" title>ComfyUI\u79d2\u51fa\u56fe-SDXL-Turbo &#8211; \u77e5\u4e4e (zhihu.com)<\/a><\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_670221383_0\">SDXL-Turbo\u6a21\u578b\u4ecb\u7ecd\uff08\u539f\u6587\u76f4\u8bd1\uff09<\/h2>\n\n\n\n<figure class=\"wp-block-image\"><noscript><img decoding=\"async\" src=\"https:\/\/pic3.zhimg.com\/80\/v2-56fc45006092c1d414b41237e6bff37a_720w.webp\" alt><\/noscript><img decoding=\"async\" src=\"data:image\/gif;base64,R0lGODlhAQABAIAAAAAAAP\/\/\/yH5BAEAAAAALAAAAAABAAEAAAIBRAA7\" alt data-src=\"https:\/\/pic3.zhimg.com\/80\/v2-56fc45006092c1d414b41237e6bff37a_720w.webp\" class=\" lazyload\"><\/figure>\n\n\n\n<p>SDXL-Turbo \u662f\u4e00\u79cd\u5feb\u901f\u6587\u672c\u5230\u56fe\u50cf\u751f\u6210\u6a21\u578b\uff0c\u53ef\u5728\u5355\u6b21\u7f51\u7edc\u8bc4\u4f30\u4e2d\u6839\u636e\u6587\u672c\u63d0\u793a\u5408\u6210\u903c\u771f\u7684\u56fe\u50cf\u3002\u5b9e\u65f6\u6f14\u793a\u8bf7\u70b9\u51fb\uff1a<a href=\"http:\/\/clipdrop.co\/stable-diffusion-turbo\" target=\"_blank\" rel=\"noopener\" title>http:\/\/clipdrop.co\/stable-diffusion-turbo<\/a><\/p>\n\n\n\n<p>SDXL-Turbo is a fast generative text-to-image model that can synthesize photorealistic images from a text prompt in a single network evaluation. A real-time demo is available here:&nbsp;<a href=\"http:\/\/clipdrop.co\/stable-diffusion-turbo\" target=\"_blank\" rel=\"noopener\" title>http:\/\/clipdrop.co\/stable-diffusion-turbo<\/a><\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"h_670221383_1\">\u6a21\u578b\u63cf\u8ff0<\/h3>\n\n\n\n<p>SDXL-Turbo \u662f&nbsp;<a href=\"https:\/\/huggingface.co\/stabilityai\/stable-diffusion-xl-base-1.0\" target=\"_blank\" rel=\"noopener\" title>SDXL 1.0<\/a>&nbsp;\u7684\u63d0\u70bc\u7248\u672c\uff0c\u7ecf\u8fc7\u8bad\u7ec3\u53ef\u7528\u4e8e\u5b9e\u65f6\u5408\u6210\u3002SDXL-Turbo \u57fa\u4e8e\u4e00\u79cd\u540d\u4e3a &#8220;\u5bf9\u6297\u6269\u6563\u63d0\u70bc&#8221;\uff08Adversarial Diffusion Distillation\uff0c\u7b80\u79f0 ADD\uff09\u7684\u65b0\u578b\u8bad\u7ec3\u65b9\u6cd5\uff08\u53c2\u89c1<a href=\"https:\/\/stability.ai\/research\/adversarial-diffusion-distillation\" target=\"_blank\" rel=\"noopener\" title>\u6280\u672f\u62a5\u544a<\/a>\uff09\uff0c\u53ef\u5728 1 \u5230 4 \u4e2a\u6b65\u9aa4\u5185\u4ee5\u9ad8\u56fe\u50cf\u8d28\u91cf\u5bf9\u5927\u89c4\u6a21\u57fa\u7840\u56fe\u50cf\u6269\u6563\u6a21\u578b\u8fdb\u884c\u91c7\u6837\u3002\u8fd9\u79cd\u65b9\u6cd5\u4f7f\u7528\u5206\u6570\u84b8\u998f\u6765\u5229\u7528\u5927\u89c4\u6a21\u73b0\u6210\u7684\u56fe\u50cf\u6269\u6563\u6a21\u578b\u4f5c\u4e3a\u6559\u5e08\u4fe1\u53f7\uff0c\u5e76\u5c06\u5176\u4e0e\u5bf9\u6297\u635f\u5931\u76f8\u7ed3\u5408\uff0c\u4ee5\u786e\u4fdd\u5373\u4f7f\u5728 1 \u6216 2 \u6b65\u91c7\u6837\u7684\u4f4e\u6b65\u9aa4\u673a\u5236\u4e0b\u4e5f\u80fd\u83b7\u5f97\u9ad8\u56fe\u50cf\u4fdd\u771f\u5ea6\u3002<\/p>\n\n\n\n<p>SDXL-Turbo is a distilled version of SDXL 1.0, trained for real-time synthesis. SDXL-Turbo is based on a novel training method called Adversarial Diffusion Distillation (ADD) (see the technical report), which allows sampling large-scale foundational image diffusion models in 1 to 4 steps at high image quality. This approach uses score distillation to leverage large-scale off-the-shelf image diffusion models as a teacher signal and combines this with an adversarial loss to ensure high image fidelity even in the low-step regime of one or two sampling steps.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"h_670221383_2\">\u6a21\u578b\u6765\u6e90<\/h3>\n\n\n\n<p>\u51fa\u4e8e\u7814\u7a76\u76ee\u7684\uff0c\u6211\u4eec\u63a8\u8350\u4f7f\u7528\u6211\u4eec\u7684\u751f\u6210\u6a21\u578b Github \u5b58\u50a8\u5e93 (<a href=\"https:\/\/github.com\/Stability-AI\/generative-models\" target=\"_blank\" rel=\"noopener\" title>https:\/\/github.com\/Stability-AI\/generative-models<\/a>)\uff0c\u5b83\u5b9e\u73b0\u4e86\u6700\u6d41\u884c\u7684\u6269\u6563\u6846\u67b6\uff08\u5305\u62ec\u8bad\u7ec3\u548c\u63a8\u7406\uff09\u3002<\/p>\n\n\n\n<p>For research purposes, we recommend our generative-models Github repository (<a href=\"https:\/\/github.com\/Stability-AI\/generative-models\" target=\"_blank\" rel=\"noopener\" title=\"https:\/\/github.com\/Stability-AI\/generative-models\">https:\/\/github.com\/Stability-AI\/generative-models<\/a>), which implements the most popular diffusion frameworks (both training and inference).<\/p>\n\n\n\n<p>\u8d44\u6e90\u5e93\uff1a<a href=\"https:\/\/github.com\/Stability-AI\/generative-models\" target=\"_blank\" rel=\"noopener\" title=\"https:\/\/github.com\/Stability-AI\/generative-models\">https:\/\/github.com\/Stability-AI\/generative-models<\/a>&nbsp;\u8bba\u6587\uff1a&nbsp;<a href=\"https:\/\/stability.ai\/research\/adversarial-diffusion-distillation\" target=\"_blank\" rel=\"noopener\" title=\"https:\/\/stability.ai\/research\/adversarial-diffusion-distillation\">https:\/\/stability.ai\/research\/adversarial-diffusion-distillation<\/a>&nbsp;\u6f14\u793a\uff1a<a href=\"http:\/\/clipdrop.co\/stable-diffusion-turbo\" target=\"_blank\" rel=\"noopener\" title=\"http:\/\/clipdrop.co\/stable-diffusion-turbo\">http:\/\/clipdrop.co\/stable-diffusion-turbo<\/a><\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"h_670221383_3\">\u8bc4\u4f30<\/h3>\n\n\n\n<figure class=\"wp-block-image\"><noscript><img decoding=\"async\" src=\"https:\/\/pic3.zhimg.com\/80\/v2-e1656cc6e58fd383dbd1dcb0917d317a_720w.webp\" alt><\/noscript><img decoding=\"async\" src=\"data:image\/gif;base64,R0lGODlhAQABAIAAAAAAAP\/\/\/yH5BAEAAAAALAAAAAABAAEAAAIBRAA7\" alt data-src=\"https:\/\/pic3.zhimg.com\/80\/v2-e1656cc6e58fd383dbd1dcb0917d317a_720w.webp\" class=\" lazyload\"><\/figure>\n\n\n\n<figure class=\"wp-block-image\"><noscript><img decoding=\"async\" src=\"https:\/\/pic4.zhimg.com\/80\/v2-232b3246f9445be075bf1b596f5f6097_720w.webp\" alt><\/noscript><img decoding=\"async\" src=\"data:image\/gif;base64,R0lGODlhAQABAIAAAAAAAP\/\/\/yH5BAEAAAAALAAAAAABAAEAAAIBRAA7\" alt data-src=\"https:\/\/pic4.zhimg.com\/80\/v2-232b3246f9445be075bf1b596f5f6097_720w.webp\" class=\" lazyload\"><\/figure>\n\n\n\n<p>\u4e0a\u8ff0\u56fe\u8868\u8bc4\u4f30\u4e86\u7528\u6237\u5bf9 SDXL-Turbo \u7684\u504f\u597d\u7a0b\u5ea6\uff0c\u800c\u4e0d\u662f\u5bf9\u5176\u4ed6\u5355\u6b65\u548c\u591a\u6b65\u578b\u53f7\u7684\u504f\u597d\u7a0b\u5ea6\u3002\u5728\u56fe\u50cf\u8d28\u91cf\u548c\u63d0\u793a\u8ddf\u8e2a\u65b9\u9762\uff0c\u5355\u6b65\u8bc4\u4f30\u7684 SDXL-Turbo \u6bd4\u56db\u6b65\uff08\u6216\u66f4\u5c11\uff09\u8bc4\u4f30\u7684 LCM-XL \u66f4\u53d7\u4eba\u7c7b\u6295\u7968\u8005\u7684\u9752\u7750\u3002\u6b64\u5916\uff0c\u6211\u4eec\u8fd8\u53d1\u73b0\uff0cSDXL-Turbo \u91c7\u7528\u56db\u4e2a\u6b65\u9aa4\u53ef\u8fdb\u4e00\u6b65\u63d0\u9ad8\u6027\u80fd\u3002\u6709\u5173\u7528\u6237\u7814\u7a76\u7684\u8be6\u7ec6\u4fe1\u606f\uff0c\u8bf7\u53c2\u9605\u7814\u7a76<a href=\"https:\/\/stability.ai\/research\/adversarial-diffusion-distillation\" target=\"_blank\" rel=\"noopener\" title>\u8bba\u6587<\/a>\u3002<\/p>\n\n\n\n<p>The charts above evaluate user preference for SDXL-Turbo over other single- and multi-step models. SDXL-Turbo evaluated at a single step is preferred by human voters in terms of image quality and prompt following over LCM-XL evaluated at four (or fewer) steps. In addition, we see that using four steps for SDXL-Turbo further improves performance. For details on the user study, we refer to the research paper.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"h_670221383_4\">\u76f4\u63a5\u4f7f\u7528<\/h3>\n\n\n\n<p>\u8be5\u6a21\u578b\u4ec5\u7528\u4e8e\u7814\u7a76\u76ee\u7684\u3002\u53ef\u80fd\u7684\u7814\u7a76\u9886\u57df\u548c\u4efb\u52a1\u5305\u62ec &#8211; \u751f\u6210\u6a21\u578b\u7814\u7a76\u3002 &#8211; \u7814\u7a76\u751f\u6210\u6a21\u578b\u7684\u5b9e\u65f6\u5e94\u7528\u3002 &#8211; \u7814\u7a76\u5b9e\u65f6\u751f\u6210\u6a21\u578b\u7684\u5f71\u54cd\u3002 &#8211; \u5b89\u5168\u90e8\u7f72\u6709\u53ef\u80fd\u751f\u6210\u6709\u5bb3\u5185\u5bb9\u7684\u6a21\u578b\u3002 &#8211; \u63a2\u7d22\u548c\u4e86\u89e3\u751f\u6210\u6a21\u578b\u7684\u5c40\u9650\u6027\u548c\u504f\u5dee\u3002 &#8211; \u827a\u672f\u4f5c\u54c1\u7684\u751f\u6210\u4ee5\u53ca\u5728\u8bbe\u8ba1\u548c\u5176\u4ed6\u827a\u672f\u8fc7\u7a0b\u4e2d\u7684\u5e94\u7528\u3002 &#8211; \u6559\u80b2\u6216\u521b\u9020\u6027\u5de5\u5177\u4e2d\u7684\u5e94\u7528\u3002 \u9664\u5916\u7528\u9014\u8bf4\u660e\u5982\u4e0b\u3002<\/p>\n\n\n\n<p>The model is intended for research purposes only. Possible research areas and tasks include &#8211; Research on generative models. &#8211; Research on real-time applications of generative models. &#8211; Research on the impact of real-time generative models. &#8211; Safe deployment of models which have the potential to generate harmful content. &#8211; Probing and understanding the limitations and biases of generative models. &#8211; Generation of artworks and use in design and other artistic processes. &#8211; Applications in educational or creative tools. Excluded uses are described below.<\/p>\n\n\n\n<p><strong>Python\u4f7f\u7528diffusers<\/strong><\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>pip install diffusers transformers accelerate --upgrade<\/code><\/pre>\n\n\n\n<p><strong>\u6587\u672c\u5230\u56fe\u50cf\uff1a<\/strong>&nbsp;SDXL-Turbo \u4e0d\u4f7f\u7528 guidance_scale \u6216 negative_prompt\uff0c\u6211\u4eec\u4f7f\u7528 guidance_scale=0.0 \u5c06\u5176\u7981\u7528\u3002\u6a21\u578b\u6700\u597d\u751f\u6210 512&#215;512 \u5c3a\u5bf8\u7684\u56fe\u50cf\uff0c\u4f46\u66f4\u5927\u5c3a\u5bf8\u7684\u56fe\u50cf\u4e5f\u53ef\u4ee5\u4f7f\u7528\u3002\u4e00\u4e2a\u6b65\u9aa4\u5c31\u8db3\u4ee5\u751f\u6210\u9ad8\u8d28\u91cf\u7684\u56fe\u50cf\u3002<\/p>\n\n\n\n<p>SDXL-Turbo does not make use of guidance_scale or negative_prompt, we disable it with guidance_scale=0.0. Preferably, the model generates images of size 512&#215;512 but higher image sizes work as well. A single step is enough to generate high quality images.<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>from diffusers import AutoPipelineForText2Image\nimport torch\n\npipe = AutoPipelineForText2Image.from_pretrained(\"stabilityai\/sdxl-turbo\", torch_dtype=torch.float16, variant=\"fp16\")\npipe.to(\"cuda\")\n\nprompt = \"\u4e00\u53ea\u7a7f\u7740\u7cbe\u81f4\u610f\u5927\u5229\u7267\u5e08\u888d\u7684\u5c0f\u6d63\u718a\u7684\u7535\u5f71\u822c\u955c\u5934\u3002\"\n\nimage = pipe(prompt=prompt, num_inference_steps=1, guidance_scale=0.0).images&#91;0]<\/code><\/pre>\n\n\n\n<p><strong>\u56fe\u50cf\u5230\u56fe\u50cf\uff1a<\/strong>&nbsp;\u4f7f\u7528 SDXL-Turbo \u751f\u6210\u56fe\u50cf\u5230\u56fe\u50cf\u65f6\uff0c\u8bf7\u786e\u4fdd num_inference_steps * strength \u5927\u4e8e\u6216\u7b49\u4e8e 1\u3002 \u56fe\u50cf\u5230\u56fe\u50cf\u6d41\u6c34\u7ebf\u5c06\u4ee5 int(num_inference_steps * strength) \u6b65\u957f\u8fd0\u884c\uff0c\u4f8b\u5982\uff0c\u5728\u4e0b\u9762\u7684\u793a\u4f8b\u4e2d\uff0c0.5 * 2.0 = 1 \u6b65\u3002\u4f7f\u7528 SDXL-Turbo \u751f\u6210\u56fe\u50cf\u5230\u56fe\u50cf\u65f6\uff0c\u8bf7\u786e\u4fdd num_inference_steps * strength \u5927\u4e8e\u6216\u7b49\u4e8e 1\u3002 \u56fe\u50cf\u5230\u56fe\u50cf\u6d41\u6c34\u7ebf\u5c06\u4ee5 int(num_inference_steps * strength) \u6b65\u957f\u8fd0\u884c\uff0c\u4f8b\u5982\uff0c\u5728\u4e0b\u9762\u7684\u793a\u4f8b\u4e2d\uff0c0.5 * 2.0 = 1 \u6b65\u3002<\/p>\n\n\n\n<p>When using SDXL-Turbo for image-to-image generation, make sure that num_inference_steps * strength is larger or equal to 1. The image-to-image pipeline will run for int(num_inference_steps * strength) steps, e.g. 0.5 * 2.0 = 1 step in our example below.<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>from diffusers import AutoPipelineForImage2Image\nfrom diffusers.utils import load_image\n\npipe = AutoPipelineForImage2Image.from_pretrained(\"stabilityai\/sdxl-turbo\", torch_dtype=torch.float16, variant=\"fp16\")\n\ninit_image = load_image(\"https:\/\/huggingface.co\/datasets\/huggingface\/documentation-images\/resolve\/main\/diffusers\/cat.png\").resize((512, 512))\n\nprompt = \"\u732b\u5deb\u5e08\uff0c\u7518\u9053\u592b\uff0c\u9b54\u6212\uff0c\u8be6\u7ec6\uff0c\u5947\u5e7b\uff0c\u53ef\u7231\uff0c\u8ff7\u4eba\uff0c\u76ae\u514b\u65af\uff0c\u8fea\u58eb\u5c3c\uff0c8k\"\n\nimage = pipe(prompt, image=init_image, num_inference_steps=2, strength=0.5, guidance_scale=0.0).images&#91;0]<\/code><\/pre>\n\n\n\n<p><strong>\u8d85\u51fa\u8303\u56f4\u7684\u4f7f\u7528<\/strong>&nbsp;\u8be5\u6a21\u578b\u672a\u7ecf\u8fc7\u57f9\u8bad\uff0c\u4e0d\u80fd\u4f5c\u4e3a\u4eba\u7269\u6216\u4e8b\u4ef6\u7684\u771f\u5b9e\u8868\u73b0\uff0c\u56e0\u6b64\u4f7f\u7528\u8be5\u6a21\u578b\u751f\u6210\u6b64\u7c7b\u5185\u5bb9\u8d85\u51fa\u4e86\u8be5\u6a21\u578b\u7684\u80fd\u529b\u8303\u56f4\u3002\u8bf7\u52ff\u4ee5\u8fdd\u53cd Stability AI \u53ef\u63a5\u53d7\u4f7f\u7528<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/stability.ai\/use-policy\" target=\"_blank\" rel=\"noreferrer noopener\">\u653f\u7b56<\/a>\u7684\u65b9\u5f0f\u4f7f\u7528\u8be5\u6a21\u578b\u3002<\/p>\n\n\n\n<p>The model was not trained to be factual or true representations of people or events, and therefore using the model to generate such content is out-of-scope for the abilities of this model. The model should not be used in any way that violates Stability AI&#8217;s Acceptable Use Policy.<\/p>\n\n\n\n<h4 class=\"wp-block-heading\" id=\"h_670221383_5\">\u9650\u5236<\/h4>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u751f\u6210\u7684\u56fe\u50cf\u5177\u6709\u56fa\u5b9a\u5206\u8fa8\u7387\uff08512&#215;512\u50cf\u7d20\uff09\uff0c\u6a21\u578b\u65e0\u6cd5\u5b9e\u73b0\u5b8c\u7f8e\u7684\u7167\u7247\u903c\u771f\u6548\u679c\u3002<\/li>\n\n\n\n<li>\u6a21\u578b\u65e0\u6cd5\u6e32\u67d3\u53ef\u8fa8\u8ba4\u7684\u6587\u672c\u3002<\/li>\n\n\n\n<li>\u751f\u6210\u7684\u4eba\u8138\u548c\u4eba\u7269\u53ef\u80fd\u4e0d\u591f\u51c6\u786e\u3002<\/li>\n\n\n\n<li>\u6a21\u578b\u7684\u81ea\u7f16\u7801\u90e8\u5206\u662f\u6709\u635f\u7684\u3002<\/li>\n<\/ul>\n\n\n\n<h4 class=\"wp-block-heading\">limitations<\/h4>\n\n\n\n<ul class=\"wp-block-list\">\n<li>The quality and prompt alignment is lower than that of&nbsp;<a href=\"https:\/\/huggingface.co\/stabilityai\/sdxl-turbo\/\" target=\"_blank\" rel=\"noopener\" title>SDXL-Turbo<\/a>.<\/li>\n\n\n\n<li>The generated images are of a fixed resolution (512&#215;512 pix), and the model does not achieve perfect photorealism.<\/li>\n\n\n\n<li>The model cannot render legible text.<\/li>\n\n\n\n<li>Faces and people in general may not be generated properly.<\/li>\n\n\n\n<li>The autoencoding part of the model is lossy.<\/li>\n<\/ul>\n\n\n\n<p><strong>\u5efa\u8bae<\/strong>&nbsp;\u8be5\u6a21\u578b\u4ec5\u7528\u4e8e\u7814\u7a76\u76ee\u7684\u3002<\/p>\n\n\n\n<p>The model is intended for research purposes only.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"h_670221383_7\">ComyfyUI\u5b89\u88c5\u548c\u4f7f\u7528<\/h3>\n\n\n\n<p>\u5df2\u5b89\u88c5\u8fc7comfyUI\u7684\u8bf7\u66f4\u65b0\u7248\u672c\uff0c\u6b64\u6a21\u578b\u9700\u8981\u4f7f\u7528\u6700\u65b0\u7684ComfyUI\u7248\u672c\uff0c\u4e0b\u8f7d\u5982\u4e0b\u8d44\u6e90\u5b8c\u6210\u540e\u89e3\u538b\u5373\u53ef\u4f7f\u7528\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5b98\u7f51\u4e0b\u8f7d\u5730\u5740\uff1a<a href=\"https:\/\/github.com\/comfyanonymous\/ComfyUI\/releases\/download\/latest\/ComfyUI_windows_portable_nvidia_cu121_or_cpu.7z\" target=\"_blank\" rel=\"noopener\" title=\"https:\/\/github.com\/comfyanonymous\/ComfyUI\/releases\/download\/latest\/ComfyUI_windows_portable_nvidia_cu121_or_cpu.7z\">https:\/\/github.com\/comfyanonymous\/ComfyUI\/releases\/download\/latest\/ComfyUI_windows_portable_nvidia_cu121_or_cpu.7z<\/a><\/li>\n\n\n\n<li>\u767e\u5ea6\u4e91\u76d8\u94fe\u63a5\uff1a&nbsp;<a href=\"https:\/\/pan.baidu.com\/s\/1rQ3J2rCh9zsjxUxJ4LDmlA?pwd=n2i7\" target=\"_blank\" rel=\"noopener\" title=\"https:\/\/pan.baidu.com\/s\/1rQ3J2rCh9zsjxUxJ4LDmlA?pwd=n2i7\">https:\/\/pan.baidu.com\/s\/1rQ3J2rCh9zsjxUxJ4LDmlA?pwd=n2i7<\/a>&nbsp;\u63d0\u53d6\u7801\uff1an2i7 \uff082023.12.01\u66f4\u65b0\uff09<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"h_670221383_8\">\u5b89\u88c5\u6a21\u578b<\/h3>\n\n\n\n<p>\u6a21\u578b\u4e0b\u8f7d\u5730\u5740\u5982\u4e0b\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5b98\u7f51\u4e0b\u8f7d\uff1a&nbsp;<a href=\"https:\/\/huggingface.co\/stabilityai\/sdxl-turbo\/tree\/main\" target=\"_blank\" rel=\"noopener\" title>https:\/\/huggingface.co\/stabilityai\/sdxl-turbo\/tree\/main<\/a><\/li>\n\n\n\n<li>\u5938\u514b\u7f51\u76d8\uff1a&nbsp;<a href=\"https:\/\/pan.quark.cn\/s\/85712b7f8ff7\" target=\"_blank\" rel=\"noopener\" title=\"https:\/\/pan.quark.cn\/s\/85712b7f8ff7\">https:\/\/pan.quark.cn\/s\/85712b7f8ff7<\/a>&nbsp;\u63d0\u53d6\u7801\uff1aariq<\/li>\n<\/ul>\n\n\n\n<p>\u4e0b\u8f7d\u5b8c\u6210\u540e\u5c06\u6a21\u578b\u5b58\u653e\u5728<code>F:\\ComfyUI_windows_portable_nightly_pytorch\\ComfyUI\\models\\checkpoints<\/code>&nbsp;\u4e0b<\/p>\n\n\n\n<figure class=\"wp-block-image\"><noscript><img decoding=\"async\" src=\"https:\/\/pic2.zhimg.com\/80\/v2-56e661ba4bd77190dc9386f81098537d_720w.webp\" alt><\/noscript><img decoding=\"async\" src=\"data:image\/gif;base64,R0lGODlhAQABAIAAAAAAAP\/\/\/yH5BAEAAAAALAAAAAABAAEAAAIBRAA7\" alt data-src=\"https:\/\/pic2.zhimg.com\/80\/v2-56e661ba4bd77190dc9386f81098537d_720w.webp\" class=\" lazyload\"><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"h_670221383_9\">\u542f\u52a8ComfyUI<\/h3>\n\n\n\n<p>\u6839\u636e\u5e73\u53f0\u542f\u52a8\uff0c\u5982\u679c\u662f\u82f1\u4f1f\u8fbe\u7684\u663e\u5361\u9009<strong>run_nvidia_gpu.bat<\/strong>&nbsp;\u5982\u679c\u6ca1\u6709\u72ec\u663e\u4f7f\u7528&nbsp;<strong>run_cpu.bat<\/strong>\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image\"><noscript><img decoding=\"async\" src=\"https:\/\/pic2.zhimg.com\/80\/v2-6671d718ce86a5441b25d9ba9fd1e649_720w.webp\" alt><\/noscript><img decoding=\"async\" src=\"data:image\/gif;base64,R0lGODlhAQABAIAAAAAAAP\/\/\/yH5BAEAAAAALAAAAAABAAEAAAIBRAA7\" alt data-src=\"https:\/\/pic2.zhimg.com\/80\/v2-6671d718ce86a5441b25d9ba9fd1e649_720w.webp\" class=\" lazyload\"><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"h_670221383_10\">\u5de5\u4f5c\u6d41\u4ecb\u7ecd<\/h3>\n\n\n\n<p>\u5de5\u4f5c\u6d41\u4e0b\u8f7d\u5730\u5740\uff1a&nbsp;<a href=\"https:\/\/pan.baidu.com\/s\/1GDfOgxii4WDZtLZHMA0o7A?pwd=e5rf\" target=\"_blank\" rel=\"noopener\" title>https:\/\/pan.baidu.com\/s\/1GDfOgxii4WDZtLZHMA0o7A?pwd=e5rf<\/a>&nbsp;\u63d0\u53d6\u7801\uff1ae5rf<\/p>\n\n\n\n<figure class=\"wp-block-image\"><noscript><img decoding=\"async\" src=\"https:\/\/pic2.zhimg.com\/80\/v2-8bb6dc7c2e20f682a8909e3e2a7f0eb1_720w.webp\" alt><\/noscript><img decoding=\"async\" src=\"data:image\/gif;base64,R0lGODlhAQABAIAAAAAAAP\/\/\/yH5BAEAAAAALAAAAAABAAEAAAIBRAA7\" alt data-src=\"https:\/\/pic2.zhimg.com\/80\/v2-8bb6dc7c2e20f682a8909e3e2a7f0eb1_720w.webp\" class=\" lazyload\"><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"h_670221383_11\">GPU\u51fa\u56fe\u901f\u5ea6<\/h3>\n\n\n\n<p>\u5355\u5f20\u51fa\u56fe\u901f\u5ea6\u4e3a0.5\u79d2\u5de6\u53f3\u7b14\u8005\u7684\u6d4b\u8bd5\u663e\u5361\u662f2080ti\uff0c\u63d0\u793a\u8bcd\u7684\u53d8\u5316\u548c\u964d\u566a\u7b97\u6cd5\u53ef\u80fd\u4f1a\u5f71\u54cd\u51fa\u56fe\u901f\u5ea6\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image\"><noscript><img decoding=\"async\" src=\"https:\/\/pic3.zhimg.com\/80\/v2-cd996669e8eded0f2f65e5f34cbe083e_720w.webp\" alt><\/noscript><img decoding=\"async\" src=\"data:image\/gif;base64,R0lGODlhAQABAIAAAAAAAP\/\/\/yH5BAEAAAAALAAAAAABAAEAAAIBRAA7\" alt data-src=\"https:\/\/pic3.zhimg.com\/80\/v2-cd996669e8eded0f2f65e5f34cbe083e_720w.webp\" class=\" lazyload\"><\/figure>\n\n\n\n<p>16\u5f20\u56fe\u901f\u5ea6\u4e3a4.9\u79d2\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image\"><noscript><img decoding=\"async\" src=\"https:\/\/pic2.zhimg.com\/80\/v2-b81fa184c7af4c970bb3e996c6eda0e1_720w.webp\" alt><\/noscript><img decoding=\"async\" src=\"data:image\/gif;base64,R0lGODlhAQABAIAAAAAAAP\/\/\/yH5BAEAAAAALAAAAAABAAEAAAIBRAA7\" alt data-src=\"https:\/\/pic2.zhimg.com\/80\/v2-b81fa184c7af4c970bb3e996c6eda0e1_720w.webp\" class=\" lazyload\"><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"h_670221383_12\">CPU\u51fa\u56fe\u901f\u5ea6<\/h3>\n\n\n\n<p>\u7ed3\u8bba\u662f\u6ca1\u6709GPU\u7684\u670b\u53cb\u4eec\u4e5f\u53ef\u4ee5\u4f53\u9a8c\u4e0bAI\u51fa\u56fe\uff0c\u5982\u4e0b\u4f7f\u7528CPU\uff08i7-11700\uff09\u8ba1\u7b97\uff0c20\u79d2\u4ee5\u5185\u80fd\u51fa\u56fe\uff0c\u867d\u7136\u76f8\u6bd4GPU\u6162\u5f88\u591a\u76f8\u6bd4\u4e4b\u524d\u5df2\u7ecf\u8fdb\u6b65\u5f88\u591a\uff0c\u53ef\u4ee5\u8bd5\u73a9\u4e86\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image\"><noscript><img decoding=\"async\" src=\"https:\/\/pic4.zhimg.com\/80\/v2-0da931ec2715f44f163392fe54b1f60b_720w.webp\" alt><\/noscript><img decoding=\"async\" src=\"data:image\/gif;base64,R0lGODlhAQABAIAAAAAAAP\/\/\/yH5BAEAAAAALAAAAAABAAEAAAIBRAA7\" alt data-src=\"https:\/\/pic4.zhimg.com\/80\/v2-0da931ec2715f44f163392fe54b1f60b_720w.webp\" class=\" lazyload\"><\/figure>\n\n\n\n<p>\u63d0\u793a\u8bcd\u7684\u53d8\u5316\u548c\u964d\u566a\u7b97\u6cd5\u53ef\u80fd\u4f1a\u5f71\u54cd\u51fa\u56fe\u901f\u5ea6\uff0c\u7b14\u8005\u6d4b\u8bd520\u79d2\u5185\u90fd\u80fd\u51fa\u56fe\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image\"><noscript><img decoding=\"async\" src=\"https:\/\/pic4.zhimg.com\/80\/v2-12083dfee31aa20b7c549794cc4fb22b_720w.webp\" alt><\/noscript><img decoding=\"async\" src=\"data:image\/gif;base64,R0lGODlhAQABAIAAAAAAAP\/\/\/yH5BAEAAAAALAAAAAABAAEAAAIBRAA7\" alt data-src=\"https:\/\/pic4.zhimg.com\/80\/v2-12083dfee31aa20b7c549794cc4fb22b_720w.webp\" class=\" lazyload\"><\/figure>\n\n\n\n<p>16\u5f20\u56fe\u901f\u5ea6195.95\u79d2<\/p>\n\n\n\n<figure class=\"wp-block-image\"><noscript><img decoding=\"async\" src=\"https:\/\/pic3.zhimg.com\/80\/v2-dbd3363d2ef3eef49150e4fd1ca2251a_720w.webp\" alt><\/noscript><img decoding=\"async\" src=\"data:image\/gif;base64,R0lGODlhAQABAIAAAAAAAP\/\/\/yH5BAEAAAAALAAAAAABAAEAAAIBRAA7\" alt data-src=\"https:\/\/pic3.zhimg.com\/80\/v2-dbd3363d2ef3eef49150e4fd1ca2251a_720w.webp\" class=\" lazyload\"><\/figure>\n\n\n\n<p>\u53e6\u5916\u4e3a\u4ec0\u4e48SDXL-Turbo\u80fd\u5728\u8fd9\u4e48\u77ed\u7684\u65f6\u95f4\u5185\u51fa\u56fe\u4e86\uff1f \u4e3b\u8981\u539f\u56e0\u8fd8\u662f\u6b64\u6a21\u578b\u964d\u566a\u53ea\u9700&nbsp;<strong>1-3\u6b21<\/strong>\u5c31\u80fd\u51fa\u56fe\uff0c\u7b14\u8005\u4e4b\u524d\u7528\u5230\u7684\u6a21\u578b\u964d\u566a\u4e00\u822c\u4f7f\u752815-30\u6b21\u3002<\/p>\n\n\n\n<p>\u76f8\u5173\u94fe\u63a5<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><a href=\"https:\/\/huggingface.co\/stabilityai\/sd-turbo\" target=\"_blank\" rel=\"noopener\" title>stabilityai\/sd-turbo \u00b7 Hugging Face<\/a><\/li>\n\n\n\n<li><a href=\"https:\/\/github.com\/Stability-AI\/stablediffusion\" target=\"_blank\" rel=\"noopener\" title=\"https:\/\/github.com\/Stability-AI\/stablediffusion\">https:\/\/github.com\/Stability-AI\/stablediffusion<\/a><\/li>\n\n\n\n<li><a href=\"https:\/\/github.com\/comfyanonymous\/ComfyUI\" target=\"_blank\" rel=\"noopener\" title=\"https:\/\/github.com\/comfyanonymous\/ComfyUI\">https:\/\/github.com\/comfyanonymous\/ComfyUI<\/a><\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>\u539f\u6587\uff1aComfyUI\u79d2\u51fa\u56fe-SDXL-Turbo &#8211; \u77e5\u4e4e (zhihu.com) SDXL-Tu [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_import_markdown_pro_load_document_selector":0,"_import_markdown_pro_submit_text_textarea":"","footnotes":""},"categories":[5,14],"tags":[73,50,69,60],"class_list":["post-3370","post","type-post","status-publish","format-standard","hentry","category-ai-learn","category-teacher","tag-ai","tag-building","tag-learning","tag-printer","entry"],"_links":{"self":[{"href":"https:\/\/www.xinyixx.com\/index.php\/wp-json\/wp\/v2\/posts\/3370","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.xinyixx.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.xinyixx.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.xinyixx.com\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.xinyixx.com\/index.php\/wp-json\/wp\/v2\/comments?post=3370"}],"version-history":[{"count":0,"href":"https:\/\/www.xinyixx.com\/index.php\/wp-json\/wp\/v2\/posts\/3370\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.xinyixx.com\/index.php\/wp-json\/wp\/v2\/media?parent=3370"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.xinyixx.com\/index.php\/wp-json\/wp\/v2\/categories?post=3370"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.xinyixx.com\/index.php\/wp-json\/wp\/v2\/tags?post=3370"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}