[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-139167-105":59,"doc-detail-139167-en":130},{"code":4,"msg":5,"data":6},0,"success",[7,13,18,23,28,33,38,43,48,51,55],{"id":8,"doc_module":4,"doc_module_name":9,"category_name":10,"show_sort_weight":11,"slug":12},1,"Document","Story & Novel",90,"story-novel",{"id":14,"doc_module":4,"doc_module_name":9,"category_name":15,"show_sort_weight":16,"slug":17},2,"Literature",80,"literature",{"id":19,"doc_module":4,"doc_module_name":9,"category_name":20,"show_sort_weight":21,"slug":22},4,"Exam",70,"exam",{"id":24,"doc_module":4,"doc_module_name":9,"category_name":25,"show_sort_weight":26,"slug":27},5,"Comic",60,"comic",{"id":29,"doc_module":4,"doc_module_name":9,"category_name":30,"show_sort_weight":31,"slug":32},6,"Technology",50,"technology",{"id":34,"doc_module":4,"doc_module_name":9,"category_name":35,"show_sort_weight":36,"slug":37},7,"Healthcare",40,"healthcare",{"id":39,"doc_module":4,"doc_module_name":9,"category_name":40,"show_sort_weight":41,"slug":42},8,"Research & Report",30,"research-report",{"id":44,"doc_module":4,"doc_module_name":9,"category_name":45,"show_sort_weight":46,"slug":47},9,"Religion & Spirituality",20,"religion-spirituality",{"id":46,"doc_module":4,"doc_module_name":9,"category_name":49,"show_sort_weight":46,"slug":50},"World Cup","world-cup",{"id":52,"doc_module":4,"doc_module_name":9,"category_name":53,"show_sort_weight":52,"slug":54},10,"Lifestyle","lifestyle",{"id":56,"doc_module":4,"doc_module_name":9,"category_name":57,"show_sort_weight":24,"slug":58},19,"General","general",{"code":4,"msg":60,"data":61},"ok",{"site_id":62,"language":63,"slug":64,"title":65,"keywords":66,"description":67,"schema_data":68,"social_meta":123,"head_meta":125,"extra_data":127,"updated_unix":129},105,"en","enhancing-image-layout-control-with-loss-guided-diffusion-models","Enhancing Image Layout Control with Loss-Guided Diffusion Models","","Diffusion models generate high-quality images from noise using a text prompt, but fine-grained layout control is limited because final composition depends strongly on the initial noise. Existing training-free approaches either modify cross-attention maps for selected tokens or define an attention-map loss and use its gradients to guide the latent. This paper provides an interpretation unifying both strategies as complementary and shows that combining them yields superior layout control while maintaining strong image quality. The proposed framework enables simultaneous layout adherence and fidelity improvements.",{"@graph":69,"@context":122},[70,84,105],{"@type":71,"itemListElement":72},"BreadcrumbList",[73,77,79,82],{"item":74,"name":75,"@type":76,"position":8},"https://docshare.wps.com","Home","ListItem",{"item":78,"name":9,"@type":76,"position":14},"https://docshare.wps.com/document/",{"item":80,"name":40,"@type":76,"position":81},"https://docshare.wps.com/document/research-report/",3,{"item":83,"name":65,"@type":76,"position":19},"https://docshare.wps.com/document/enhancing-image-layout-control-with-loss-guided-diffusion-models/139167/",{"url":83,"name":65,"@type":85,"image":86,"author":91,"headline":65,"publisher":94,"fileFormat":97,"inLanguage":63,"description":67,"dateModified":98,"datePublished":99,"encodingFormat":97,"isAccessibleForFree":100,"interactionStatistic":101},"DigitalDocument",{"url":87,"@type":88,"width":89,"height":90},"https://docshare.wps.com/thumbnails/enhancing-image-layout-control-with-loss-guided-diffusion-models/139167.png","ImageObject",300,407,{"name":92,"@type":93},"Lucas Vance","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-10-16","2026-08-23",true,{"@type":102,"interactionType":103,"userInteractionCount":39},"InteractionCounter",{"@type":104},"ViewAction",{"@type":106,"mainEntity":107},"FAQPage",[108,114,118],{"name":109,"@type":110,"acceptedAnswer":111},"What problem does the paper address in diffusion-based image generation?","Question",{"text":112,"@type":113},"It addresses the lack of fine-grained control over image layout when using text prompts, since the final layout is highly dependent on the initial noise sample.","Answer",{"name":115,"@type":110,"acceptedAnswer":116},"How do previous training-free layout control methods work?",{"text":117,"@type":113},"They either directly modify cross-attention maps for specific text tokens to enhance certain regions, or they define a loss over cross-attention maps and use the loss gradient to guide the latent.",{"name":119,"@type":110,"acceptedAnswer":120},"What is injection loss guidance (iLGD) and why is it effective?",{"text":121,"@type":113},"iLGD is a training-free framework that combines attention injection and loss-guided refinement so that images achieve the desired layout while preserving image quality; the injection provides a coarse bias, and loss guidance acts as a refiner.","https://schema.org",{"og:url":83,"og:type":124,"og:title":65,"og:site_name":95,"og:description":67},"article",{"robots":126,"canonical":83},"index,follow",{"doc_id":128,"site_id":62},139167,1787505518,{"code":4,"msg":5,"data":131},{"doc_id":128,"user_id":132,"nickname":92,"user_avatar":133,"doc_module":4,"category_id":39,"category_name":40,"doc_title":65,"doc_description":67,"doc_content":134,"file_id":135,"file_url":136,"file_type":137,"file_size":138,"view_count":39,"is_deleted":4,"is_public":8,"is_downloadable":8,"audit_status":8,"page_count":44,"language":139,"language_code":63,"site_id":62,"html_lang":63,"table_of_contents":140,"faqs":141,"seo_title":142,"seo_description":67,"update_tm":129,"read_time":143},549768064622,"https://ap-avatar.wpscdn.com/davatar_6f874abed73319feea01a86fa6f0fab8","This WACV paper is the Open Access version, provided by the Computer Vision Foundation.  \nExcept for this watermark, it is identical to the accepted version; the final published version of the proceedings is available on IEEE Xplore.  \nEnhancing Image Layout Control with Loss-Guided Diffusion Models  \nZakaria Patel  \nLeapTools  \n1255 Bay St. Unit 403, Toronto, ON M5R 2A9  \n[zakaria@leaptools.com](zakaria@leaptools.com)  \nKirill Serkh University of Toronto  \n40 St. George St., Toronto, ON M5S 2E4  \n[kserkh@math.toronto.edu](kserkh@math.toronto.edu)  \nAbstract  \nDiffusion models are a powerful class of generative models capable of producing high-quality images from pure noise using a simple text prompt. While most methods which introduce additional spatial constraints into the generated images (e.g., bounding boxes) require fine-tuning, a smaller and more recent subset of these methods take advantage of the models’ attention mechanism, and are training-free. These methods generally fall into one of two categories. The first entails modifying the cross-attention maps of specific tokens directly to enhance the signal in certain regions of the image. The second works by defining a loss function over the cross-attention maps, and using the gradient of this loss to guide the latent. While previous work explores these as alternative strategies, we provide an interpretation for these methods which highlights their complimentary features, and demonstrate that it is possible to obtain superior performance when both methods are used in concert.  \n1. Introduction  \nRecently, diffusion models [11, 20, 22] have emerged asa powerful class of generative models capable of producing high quality samples with superior mode coverage. These models are trained to approximate a data distribution, and sampling from this learned distribution can subsequently generate very realistic and diverse images. Incorporating conditioning [16, 17, 18] further extends the utility of diffusion models by allowing one to specify the contents of the desired image using a simple text prompt, leveraging the models’ impressive compositional capabilities to combine concepts in ways that may not have been present in the training set. Conditioning on a text prompt does not, however, allow one fine-grained control over the layout of the final image, which is instead highly dependent on the initial noise sample.  \nAn especially simple and intuitive way of describing a layout is to provide bounding boxes for various tokens in the text prompt. One way to realize a model taking such an  \ninput is to resort to training-based methods, wherein a pretrained model undergoes additional finetuning using training data where the images have been supplemented by their layouts. While such methods can achieve impressive performance [29, 30], this often involves the introduction of additional model complexity and training cost, in addition to the difficult task of compiling the training data. A recently proposed alternative approach [5, 28] uses the cross-attention module to achieve training-free layout control.  \nIn this paper, we propose injection loss guidance (iLGD), a training-free framework in which the model’s denoising process works in synergy with loss guidance, such that sampled images simultaneously present the appropriate layout and maintain good image quality. First, we bias the latent towards the desired layout by altering the model’s attention maps directly, a process which we refer to as attention injection. The goal of injection is not to produce a latent which perfectly adheres to the desired layout, but rather to provide a comfortable starting point for loss guidance. The resulting latent is sufficiently close to the desired layout that we can afford to use a smaller amount of loss guidance. In doing so, we reimagine the role of injection as a coarse biasing of the diffusion process, and loss guidance as a refiner. We show that such a framework is capable of controlling th","cbCairJiDppve0R0","https://ap.wps.com/l/cbCairJiDppve0R0","pdf",5103108,"English","# Introduction\n# Related Work\n## Generative Models\n## Controllable Generation","[{\"question\":\"What problem does the paper address in diffusion-based image generation?\",\"answer\":\"It addresses the lack of fine-grained control over image layout when using text prompts, since the final layout is highly dependent on the initial noise sample.\"},{\"question\":\"How do previous training-free layout control methods work?\",\"answer\":\"They either directly modify cross-attention maps for specific text tokens to enhance certain regions, or they define a loss over cross-attention maps and use the loss gradient to guide the latent.\"},{\"question\":\"What is injection loss guidance (iLGD) and why is it effective?\",\"answer\":\"iLGD is a training-free framework that combines attention injection and loss-guided refinement so that images achieve the desired layout while preserving image quality; the injection provides a coarse bias, and loss guidance acts as a refiner.\"}]","Enhancing Image Layout Control with Loss-Guided Diffusion Models | PDF",23]