{"id":21938,"date":"2024-06-05T15:55:44","date_gmt":"2024-06-05T22:55:44","guid":{"rendered":"http:\/\/jnack.com\/blog\/?p=21938"},"modified":"2024-06-05T15:57:29","modified_gmt":"2024-06-05T22:57:29","slug":"hyperdreambooth-explained-in-5-minutes","status":"publish","type":"post","link":"https:\/\/jnack.com\/blog\/2024\/06\/05\/hyperdreambooth-explained-in-5-minutes\/","title":{"rendered":"HyperDreamBooth, explained in 5 minutes"},"content":{"rendered":"\n<p>My former Google teammates have been cranking out some amazing AI personalization tech, with <a href=\"https:\/\/hyperdreambooth.github.io\/\">HyperDreamBooth<\/a> far surpassing the performance of their original DreamBooth (y&#8217;know, from 2022\u2014such a simpler ancient time!). Here they offer a short &amp; pretty accessible overview of how it works:<\/p>\n\n\n\n<blockquote class=\"wp-block-quote\">\n<p>Using only a&nbsp;<em>single<\/em>&nbsp;input image,&nbsp;<em>HyperDreamBooth<\/em>&nbsp;is able to personalize a text-to-image diffusion model&nbsp;<strong>25x<\/strong>&nbsp;faster than DreamBooth, by using&nbsp;<strong>(1)<\/strong>&nbsp;a HyperNetwork to generate an initial prediction of a subset of network weights that are then&nbsp;<strong>(2)<\/strong>&nbsp;refined using fast finetuning for high fidelity to subject detail. Our method both&nbsp;<em>conserves model integrity and style diversity<\/em>&nbsp;while closely approximating the subject&#8217;s essence and details.<\/p>\n<\/blockquote>\n\n\n\n<figure class=\"wp-block-embed is-type-video is-provider-youtube wp-block-embed-youtube wp-embed-aspect-16-9 wp-has-aspect-ratio\"><div class=\"wp-block-embed__wrapper\">\n<iframe loading=\"lazy\" title=\"HyperDreamBooth CVPR 2024 Talk\" width=\"604\" height=\"340\" src=\"https:\/\/www.youtube.com\/embed\/thBgIY7ret8?feature=oembed\" frameborder=\"0\" allow=\"accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share\" referrerpolicy=\"strict-origin-when-cross-origin\" allowfullscreen><\/iframe>\n<\/div><\/figure>\n","protected":false},"excerpt":{"rendered":"<p>My former Google teammates have been cranking out some amazing AI personalization tech, with HyperDreamBooth far surpassing the performance of their original DreamBooth (y&#8217;know, from 2022\u2014such a simpler ancient time!). Here they offer a short &amp; pretty accessible overview of how it works: Using only a&nbsp;single&nbsp;input image,&nbsp;HyperDreamBooth&nbsp;is able to personalize a text-to-image diffusion model&nbsp;25x&nbsp;faster than [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":[],"categories":[66],"tags":[],"_links":{"self":[{"href":"https:\/\/jnack.com\/blog\/wp-json\/wp\/v2\/posts\/21938"}],"collection":[{"href":"https:\/\/jnack.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/jnack.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/jnack.com\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/jnack.com\/blog\/wp-json\/wp\/v2\/comments?post=21938"}],"version-history":[{"count":6,"href":"https:\/\/jnack.com\/blog\/wp-json\/wp\/v2\/posts\/21938\/revisions"}],"predecessor-version":[{"id":21946,"href":"https:\/\/jnack.com\/blog\/wp-json\/wp\/v2\/posts\/21938\/revisions\/21946"}],"wp:attachment":[{"href":"https:\/\/jnack.com\/blog\/wp-json\/wp\/v2\/media?parent=21938"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/jnack.com\/blog\/wp-json\/wp\/v2\/categories?post=21938"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/jnack.com\/blog\/wp-json\/wp\/v2\/tags?post=21938"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}