
{"id":11170,"date":"2026-08-18T15:59:04","date_gmt":"2026-08-18T07:59:04","guid":{"rendered":"https:\/\/infernews.com\/blog\/arxiv-org\/"},"modified":"2026-08-18T15:59:04","modified_gmt":"2026-08-18T07:59:04","slug":"arxiv-org","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/arxiv-org\/","title":{"rendered":"[\u6280\u8853\u6587\u7ae0] pixel-space \u6a21\u578b\u53ef\u4ee5\u8ffd\u5e73\u751a\u81f3\u8d85\u904e latent-space \u5c0d\u61c9\u6a21\u578b"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/08\/pasted-f4bd083d97d2.jpg\" alt=\"Hero image preview\"><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u4e0a\u5716\u662f\u900f\u904e pixel-space text-to-image diffusion models \u7684\u65b9\u6cd5\u8a13\u7df4\u7684 Z-Image-Turbo (Pixel) \u7522\u751f\u7684\u5716\u50cf\u3002\u5728\u4f01\u696d\u7d1a H800 GPU \u4e0a\uff0c\u63a8\u7406\u5ef6\u9072\u50c5\u7d04 0.2 \u79d2\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Pixel-space text-to-image diffusion models \u4e00\u76f4\u6709\u5169\u500b\u5438\u5f15\u4f4d\uff1a\u76f4\u63a5\u751f\u6210\u50cf\u7d20\uff0c\u5514\u53d7 VAE \u58d3\u7e2e\u4e0a\u9650\u9650\u5236\uff0c\u4ea6\u53ef\u4ee5\u7701\u53bb\u89e3\u78bc\u6b65\u9a5f\u3002\u4f46\u55ba\u5927\u898f\u6a21\u8a13\u7df4\u4e0b\uff0c\u4f62\u54cb\u6536\u6582\u901f\u5ea6\u660e\u986f\u6162\u904e latent-space \u6a21\u578b\uff0c\u4ee4\u300c\u756b\u8cea\u3001\u6548\u7387\u3001\u53ef\u8a13\u7df4\u6027\u300d\u4e09\u8005\u4e4b\u9593\u5605\u5e73\u8861\u4e00\u76f4\u5514\u5920\u6210\u719f\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5462\u7bc7\u5de5\u4f5c\u91dd\u5c0d\u5605\u6b63\u4fc2\u5462\u500b\u843d\u5dee\u3002\u4f5c\u8005\u5148\u6bd4\u8f03 pixel-space \u540c latent-space \u55ba\u5927\u898f\u6a21 pre-training \u4e0b\u5605\u884c\u70ba\uff0c\u518d\u63d0\u51fa latent-to-pixel \u7b56\u7565\uff1a\u5148\u55ba latent space \u5b78\u5230\u751f\u6210\u5148\u9a57\uff0c\u518d\u8f49\u53bb pixel space \u505a post-training\uff0c\u907f\u514d\u4e00\u958b\u59cb\u5c31\u55ba\u96e3\u5ea6\u66f4\u9ad8\u5605\u50cf\u7d20\u7a7a\u9593\u786c\u63a8\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u4f62\u54cb\u4ea6\u7cfb\u7d71\u7814\u7a76\u5497\u591a\u500b\u8f49\u63db\u8a2d\u8a08\uff0c\u5305\u62ec weight initialization\u3001data composition\u3001prediction target\u3001decoder architecture \u540c noise schedule\uff0c\u6574\u7406\u51fa\u4e00\u5957\u53ef\u884c\u505a\u6cd5\u3002\u7d50\u679c\u986f\u793a\uff0c\u505a\u6cd5\u8abf\u6574\u5f97\u7576\u4e4b\u5f8c\uff0cpixel-space \u6a21\u578b\u53ef\u4ee5\u8ffd\u5e73\u751a\u81f3\u8d85\u904e latent-space \u5c0d\u61c9\u6a21\u578b\uff0c\u63a8\u7406\u7aef\u4ef2\u6709 3.18 \u5230 4.75 \u500d\u5605\u901f\u5ea6\u63d0\u5347\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u76f4\u63a5\u8655\u7406 pixel-space diffusion models \u55ba\u5927\u898f\u6a21\u8a13\u7df4\u6642\u6536\u6582\u6162\u5605\u554f\u984c<\/li>\n<li>\u7528 latent-to-pixel \u8def\u7dda\uff0c\u5148\u5b78\u5148\u9a57\uff0c\u518d\u8f49\u5165\u50cf\u7d20\u7a7a\u9593\u5fae\u8abf<\/li>\n<li>\u6aa2\u8996\u591a\u500b\u95dc\u9375\u8a13\u7df4\u9078\u9805\uff0c\u6435\u51fa\u5be6\u7528 recipe<\/li>\n<li>\u5728\u756b\u8cea\u4e0d\u9000\u8b93\u4e4b\u4e0b\uff0c\u63db\u53d6\u66f4\u5feb\u7684 end-to-end inference<\/li>\n<li>\u5c0d\u60f3\u505a\u9ad8\u6548\u6587\u751f\u5716\u7cfb\u7d71\u3001\u53c8\u5728\u610f\u751f\u6210\u7d30\u7bc0\u4fdd\u7559\u5605\u5718\u968a\u6700\u6709\u53c3\u8003\u50f9\u503c<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/arxiv.org\/pdf\/2608.16887\" rel=\"noopener noreferrer\"><strong>Paper<\/strong><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u7814\u7a76\u805a\u7126 pixel-space text-to-image diffusion models \u7684\u8a13\u7df4\u5361\u4f4d\uff0c\u627e\u51fa\u9ede\u6a23\u55ba\u5927\u898f\u6a21\u689d\u4ef6\u4e0b\u517c\u9867\u756b\u8cea\u540c\u901f\u5ea6\u3002\u4f5c\u8005\u7528 latent-to-pixel \u7b56\u7565\uff0c\u4ee4 pixel-space \u6a21\u578b\u66f4\u63a5\u8fd1\u751a\u81f3\u8d85\u904e latent-space \u8868\u73fe\u3002<\/p>\n","protected":false},"author":8,"featured_media":11169,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"ai_generated_summary":"","footnotes":""},"categories":[171,163,165,30,129,76,127,198],"tags":[],"class_list":["post-11170","post","type-post","status-publish","format-standard","hentry","category-171","category-163","category-165","category-image","category-txt2img","category-76","category-127","category-198"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/11170","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/users\/8"}],"replies":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/comments?post=11170"}],"version-history":[{"count":0,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/11170\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/11169"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=11170"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=11170"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=11170"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}