{"id":85186,"date":"2025-08-22T14:19:20","date_gmt":"2025-08-22T14:19:20","guid":{"rendered":"https:\/\/www.newsbeep.com\/uk\/85186\/"},"modified":"2025-08-22T14:19:20","modified_gmt":"2025-08-22T14:19:20","slug":"the-tech-behind-youtube-real-time-generative-ai-effects","status":"publish","type":"post","link":"https:\/\/www.newsbeep.com\/uk\/85186\/","title":{"rendered":"The tech behind YouTube real-time generative AI effects"},"content":{"rendered":"<p>                The teacher and the student<\/p>\n<p data-block-key=\"iiwr9\">Our approach revolves around a concept called <a href=\"https:\/\/en.wikipedia.org\/wiki\/Knowledge_distillation\" target=\"_blank\" rel=\"noopener noreferrer nofollow\">knowledge distillation<\/a>, which uses a &#8220;teacher\u2013student&#8221; model training method. We start with a &#8220;teacher&#8221; \u2014 a large, powerful, pre-trained generative model that is an expert at creating the desired visual effect but is far too slow for real-time use. The type of teacher model varies depending on the goal. Initially, we used a custom-trained <a href=\"https:\/\/arxiv.org\/abs\/1912.04958\" target=\"_blank\" rel=\"noopener noreferrer nofollow\">StyleGAN2<\/a> model, which was trained on our curated dataset for real-time facial effects. This model could be paired with tools like<a href=\"https:\/\/arxiv.org\/abs\/2103.17249\" target=\"_blank\" rel=\"noopener noreferrer nofollow\"> StyleCLIP<\/a>, which allowed it to manipulate facial features based on text descriptions. This provided a strong foundation. As our project advanced, we transitioned to more sophisticated generative models like Google DeepMind\u2019s <a href=\"https:\/\/deepmind.google\/models\/imagen\/\" target=\"_blank\" rel=\"noopener noreferrer nofollow\">Imagen<\/a>. This strategic shift significantly enhanced our capabilities, enabling higher-fidelity and more diverse imagery, greater artistic control, and a broader range of styles for our on-device generative AI effects.<\/p>\n<p data-block-key=\"abdv\">The &#8220;student&#8221; is the model that ultimately runs on the user\u2019s device. It needs to be small, fast, and efficient. We designed a student model with a <a href=\"https:\/\/link.springer.com\/content\/pdf\/10.1007\/978-3-319-24574-4_28.pdf\" target=\"_blank\" rel=\"noopener noreferrer nofollow\">UNet<\/a>-based architecture, which is excellent for image-to-image tasks. It uses a <a href=\"https:\/\/research.google\/blog\/mobilenets-open-source-models-for-efficient-on-device-vision\/\" rel=\"nofollow noopener\" target=\"_blank\">MobileNet<\/a> backbone as its encoder, a design known for its performance on mobile devices, paired with a decoder that utilizes MobileNet blocks.<\/p>\n","protected":false},"excerpt":{"rendered":"The teacher and the student Our approach revolves around a concept called knowledge distillation, which uses a &#8220;teacher\u2013student&#8221;&hellip;\n","protected":false},"author":2,"featured_media":16579,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[21],"tags":[4323,86,56,54,55],"class_list":["post-85186","post","type-post","status-publish","format-standard","has-post-thumbnail","category-computing","tag-computing","tag-technology","tag-uk","tag-united-kingdom","tag-unitedkingdom"],"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/posts\/85186","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/comments?post=85186"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/posts\/85186\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/media\/16579"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/media?parent=85186"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/categories?post=85186"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/tags?post=85186"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}