[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-seo-297608-105":3,"detail-sidebar-cat-0-en-105":81,"doc-detail-297608-en":130},{"code":4,"msg":5,"data":6},0,"ok",{"site_id":7,"language":8,"slug":9,"title":10,"keywords":11,"description":12,"schema_data":13,"social_meta":74,"head_meta":76,"extra_data":78,"updated_unix":80},105,"en","diffusion-models-beat-gans-on-image-synthesis","Diffusion Models Beat GANs on Image Synthesis","","Diffusion models are shown to outperform state-of-the-art generative models for image synthesis, delivering higher image sample quality through architectural improvements and systematic ablations. Unconditional synthesis benefits from a better model architecture, while conditional synthesis uses classifier guidance to trade diversity for fidelity via gradients from a classifier. Reported results include strong FID scores across ImageNet resolutions and matching BigGAN-deep with limited forward passes, with additional gains from upsampling diffusion.",{"@graph":14,"@context":73},[15,34,56],{"@type":16,"itemListElement":17},"BreadcrumbList",[18,23,27,31],{"item":19,"name":20,"@type":21,"position":22},"https://docshare.wps.com","Home","ListItem",1,{"item":24,"name":25,"@type":21,"position":26},"https://docshare.wps.com/document/","Document",2,{"item":28,"name":29,"@type":21,"position":30},"https://docshare.wps.com/document/research-report/","Research & Report",3,{"item":32,"name":10,"@type":21,"position":33},"https://docshare.wps.com/document/diffusion-models-beat-gans-on-image-synthesis/297608/",4,{"url":32,"name":10,"@type":35,"image":36,"author":41,"headline":10,"publisher":44,"fileFormat":47,"inLanguage":8,"description":12,"dateModified":48,"datePublished":49,"encodingFormat":47,"isAccessibleForFree":50,"interactionStatistic":51},"DigitalDocument",{"url":37,"@type":38,"width":39,"height":40},"https://docshare.wps.com/thumbnails/diffusion-models-beat-gans-on-image-synthesis/297608.png","ImageObject",300,407,{"name":42,"@type":43},"Aran","Person",{"url":19,"name":45,"@type":46},"DocShare","Organization","application/pdf","2026-10-17","2026-09-18",true,{"@type":52,"interactionType":53,"userInteractionCount":55},"InteractionCounter",{"@type":54},"ViewAction",6,{"@type":57,"mainEntity":58},"FAQPage",[59,65,69],{"name":60,"@type":61,"acceptedAnswer":62},"What is the main finding about diffusion models versus GANs for image synthesis?","Question",{"text":63,"@type":64},"Diffusion models achieve superior image sample quality compared with current state-of-the-art generative models, outperforming GANs on multiple metrics and datasets.","Answer",{"name":66,"@type":61,"acceptedAnswer":67},"How do the authors improve diffusion models for unconditional image synthesis?",{"text":68,"@type":64},"They improve the model architecture through a series of ablations, achieving higher-quality unconditional samples.",{"name":70,"@type":61,"acceptedAnswer":71},"What is classifier guidance and how does it affect results?",{"text":72,"@type":64},"Classifier guidance uses gradients from a classifier during sampling, enabling a controllable trade-off between diversity and fidelity by tuning a single hyperparameter (the gradient scale).","https://schema.org",{"og:url":32,"og:type":75,"og:title":10,"og:site_name":45,"og:description":12},"article",{"robots":77,"canonical":32},"index,follow",{"doc_id":79,"site_id":7},297608,1790954726,{"code":4,"msg":82,"data":83},"success",[84,88,92,96,101,105,110,114,119,122,126],{"id":22,"doc_module":4,"doc_module_name":25,"category_name":85,"show_sort_weight":86,"slug":87},"Story & Novel",90,"story-novel",{"id":26,"doc_module":4,"doc_module_name":25,"category_name":89,"show_sort_weight":90,"slug":91},"Literature",80,"literature",{"id":33,"doc_module":4,"doc_module_name":25,"category_name":93,"show_sort_weight":94,"slug":95},"Exam",70,"exam",{"id":97,"doc_module":4,"doc_module_name":25,"category_name":98,"show_sort_weight":99,"slug":100},5,"Comic",60,"comic",{"id":55,"doc_module":4,"doc_module_name":25,"category_name":102,"show_sort_weight":103,"slug":104},"Technology",50,"technology",{"id":106,"doc_module":4,"doc_module_name":25,"category_name":107,"show_sort_weight":108,"slug":109},7,"Healthcare",40,"healthcare",{"id":111,"doc_module":4,"doc_module_name":25,"category_name":29,"show_sort_weight":112,"slug":113},8,30,"research-report",{"id":115,"doc_module":4,"doc_module_name":25,"category_name":116,"show_sort_weight":117,"slug":118},9,"Religion & Spirituality",20,"religion-spirituality",{"id":117,"doc_module":4,"doc_module_name":25,"category_name":120,"show_sort_weight":117,"slug":121},"World Cup","world-cup",{"id":123,"doc_module":4,"doc_module_name":25,"category_name":124,"show_sort_weight":123,"slug":125},10,"Lifestyle","lifestyle",{"id":127,"doc_module":4,"doc_module_name":25,"category_name":128,"show_sort_weight":97,"slug":129},19,"General","general",{"code":4,"msg":82,"data":131},{"doc_id":79,"user_id":132,"nickname":42,"user_avatar":133,"doc_module":4,"category_id":111,"category_name":29,"doc_title":10,"doc_description":12,"doc_content":134,"file_id":135,"file_url":136,"file_type":137,"file_size":138,"view_count":55,"is_deleted":4,"is_public":22,"is_downloadable":22,"audit_status":22,"page_count":139,"language":140,"language_code":8,"site_id":7,"html_lang":8,"table_of_contents":141,"faqs":142,"seo_title":143,"seo_description":12,"update_tm":144,"read_time":145,"preview_page_count":4},137455076865,"https://ap-avatar.wpscdn.com/davatar_29158cc5080c5b710cf443261637dec0","Diffusion Models Beat GANs on Image Synthesis  \nPrafulla Dhariwal􀀃  \nOpenAI [prafulla@openai.com](prafulla@openai.com)  \nAlex Nichol􀀃 OpenAI  \n[alex@openai.com](alex@openai.com)  \narXiv :2105 .05233v4 [ cs .LG] 1 Jun 2021  \nAbstract  \nWe show that diffusion models can achieve image sample quality superior to the current state-of-the-art generative models. We achieve this on unconditional image synthesis by finding a better architecture through a series of ablations. For conditional image synthesis, we further improve sample quality with classifier guidance: a simple, compute-efficient method for trading off diversity for fidelity using gradients from a classifier. We achieve an FID of 2.97 on ImageNet 128􀀂128, 4.59 on ImageNet 256􀀂256, and 7.72 on ImageNet 512􀀂512, and we match BigGAN-deep even with as few as 25 forward passes per sample, all while maintaining better coverage of the distribution. Finally, we find that classifier guidance combines well with upsampling diffusion models, further improving FID to 3.94 on ImageNet 256􀀂256 and 3.85 on ImageNet 512􀀂512 . We release our code at [https://github.com/openai/guided-diffusion](https://github.com/openai/guided-diffusion).  \n1 Introduction  \nFigure 1: Selected samples from our best ImageNet 512􀀂512 model (FID 3.85)  \nOver the past few years, generative models have gained the ability to generate human-like natural language [6], infinite high-quality synthetic images [5, 28, 51] and highly diverse human speech and music [64, 13] . These models can be used in a variety of ways, such as generating images from text prompts [72, 50] or learning useful feature representations [14, 7] . While these models are already  \n􀀃 Equal contribution  \ncapable of producing realistic images and sound, there is still much room for improvement beyond the current state-of-the-art, and better generative models could have wide-ranging impacts on graphic design, games, music production, and countless other fields.  \nGANs [19] currently hold the state-of-the-art on most image generation tasks [5, 68, 28] as measured by sample quality metrics such as FID [23], Inception Score [54] and Precision [32] . However, some of these metrics do not fully capture diversity, and it has been shown that GANs capture less diversity than state-of-the-art likelihood-based models [51, 43, 42] . Furthermore, GANs are often difficult to train, collapsing without carefully selected hyperparameters and regularizers [5, 41, 4] .  \nWhile GANs hold the state-of-the-art, their drawbacks make them difficult to scale and apply to new domains. As a result, much work has been done to achieve GAN-like sample quality with likelihood-based models [51, 25, 42, 9] . While these models capture more diversity and are typically easier to scale and train than GANs, they still fall short in terms of visual sample quality. Furthermore, except for VAEs, sampling from these models is slower than GANs in terms of wall-clock time. Diffusion models are a class of likelihood-based models which have recently been shown to produce high-quality images [56, 59, 25] while offering desirable properties such as distribution coverage, a stationary training objective, and easy scalability. These models generate samples by gradually removing noise from a signal, and their training objective can be expressed as a reweighted variational lower-bound [25] . This class of models already holds the state-of-the-art [60] on CIFAR-10 [31], but still lags behind GANs on difficult generation datasets like LSUN and ImageNet. Nichol and Dhariwal [43] found that these models improve reliably with increased compute, and can produce high-quality samples even on the difficult ImageNet 256􀀂256 dataset using an upsampling stack. However, the FID of this model is still not competitive with BigGAN-deep [5], the current state-of-the-art on this dataset.  \nWe hypothesize that the gap between diffusion models and GANs stems from at least two factors:first, that the model architectures used by re","cbCaih90qdA6kVNC","https://ap.wps.com/l/cbCaih90qdA6kVNC","pdf",3481095,44,"English","# Abstract\n# Introduction\n# Background\n# Architecture Improvements\n# Classifier Guidance for Diffusion\n# Unconditional and Conditional Results","[{\"question\":\"What is the main finding about diffusion models versus GANs for image synthesis?\",\"answer\":\"Diffusion models achieve superior image sample quality compared with current state-of-the-art generative models, outperforming GANs on multiple metrics and datasets.\"},{\"question\":\"How do the authors improve diffusion models for unconditional image synthesis?\",\"answer\":\"They improve the model architecture through a series of ablations, achieving higher-quality unconditional samples.\"},{\"question\":\"What is classifier guidance and how does it affect results?\",\"answer\":\"Classifier guidance uses gradients from a classifier during sampling, enabling a controllable trade-off between diversity and fidelity by tuning a single hyperparameter (the gradient scale).\"}]","Diffusion Models Beat GANs on Image Synthesis | PDF",1789746241,111]