北京大学、上海人工知能研究所、南洋理工大学は協働して、漫画生成AI「DiffSensei」を開発した。一般的な画像生成AIと異なり、コマ割りも行い完成した漫画原稿を生成することができると機器之心Proが報じた。 北京大学などの研究成果「漫画生成AI」 このDiffSenseiに関しては、論文「DiffSensei: Bridging Multi-Modal LLMs and Diffusion M...

<div class="article-menu-wrapper"> <div class="article-menu-wrapper-inner"> <div class="article-menu-bg"></div> <div class="gradient-transition" id="reading-progress-fill"></div> <div class="article-menu gradient-transition"> <div class="article-menu__toc flex items-center gap-1" id="article-menu__toc"> <svg clip-rule="evenodd" fill-rule="evenodd" height="24" stroke-linejoin="round" stroke-miterlimit="2" viewbox="0 0 84 98" width="20" xml:space="preserve" xmlns="http://www.w3.org/2000/svg"> <path d="m30.881 77.277 1.229.921.153-26.117V40.098c.308-4.917 4.302-8.757 9.219-8.757a9.186 9.186 0 0 1 9.218 8.91l2.611 21.355c.154 1.843-1.075 3.533-3.072 3.841-1.843.307-3.533-1.076-3.841-3.073l-2.611-21.507v-.461c0-1.229-1.076-2.304-2.304-2.304-1.229 0-2.304.921-2.304 2.304v11.675l-.153 33.184c0 1.383-.769 2.459-1.998 3.073-1.229.614-2.612.46-3.533-.308l-6.76-5.223-.153-.153c-.922-.769-4.917-3.841-8.911-1.844 1.998 2.459 5.378 6.453 9.525 10.447a3.579 3.579 0 0 1 .153 4.916c-.614.769-1.537 1.076-2.458 1.076-.922 0-1.69-.307-2.304-.921-7.528-7.068-12.291-13.673-12.444-13.981-.922-1.382-.768-3.072.307-4.301 7.219-8.451 16.436-4.15 20.431-.77v.001Zm47.011 19.664c.307 0 .614.153.768.153 1.537 0 2.919-1.076 3.38-2.611 2.611-10.446.461-21.508.461-21.969-.154-.921-.614-1.69-1.383-2.15C75.895 66.83 69.135 63.758 61.3 61.3c-1.843-.614-3.687.46-4.301 2.304-.614 1.843.461 3.687 2.304 4.301 6.605 1.998 12.291 4.455 16.745 7.374.461 3.073 1.229 10.601-.614 17.667-.615 1.691.614 3.534 2.457 3.995h.001ZM21.97 8.297a4.132 4.132 0 0 0 4.148-4.149A4.13 4.13 0 0 0 21.97 0a4.13 4.13 0 0 0-4.148 4.148 4.132 4.132 0 0 0 4.148 4.149Zm18.128 0a4.132 4.132 0 0 0 4.148-4.149A4.13 4.13 0 0 0 40.098 0a4.13 4.13 0 0 0-4.148 4.148 4.133 4.133 0 0 0 4.148 4.149Zm-35.949 0a4.132 4.132 0 0 0 4.148-4.149A4.13 4.13 0 0 0 4.149 0 4.132 4.132 0 0 0 0 4.148a4.133 4.133 0 0 0 4.149 4.149ZM21.97 25.042a4.13 4.13 0 0 0 4.148-4.148 4.132 4.132 0 0 0-4.148-4.149 4.132 4.132 0 0 0-4.148 4.149 4.13 4.13 0 0 0 4.148 4.148Zm18.128 0a4.13 4.13 0 0 0 4.148-4.148 4.132 4.132 0 0 0-4.148-4.149 4.132 4.132 0 0 0-4.149 4.149 4.133 4.133 0 0 0 4.149 4.148Zm-35.95 0a4.132 4.132 0 0 0 4.149-4.148 4.132 4.132 0 0 0-4.149-4.149A4.132 4.132 0 0 0 0 20.894a4.13 4.13 0 0 0 4.148 4.148ZM21.97 41.787a4.13 4.13 0 0 0 4.148-4.148 4.13 4.13 0 0 0-4.148-4.148 4.132 4.132 0 0 0-4.149 4.148 4.132 4.132 0 0 0 4.149 4.148Zm-17.822 0a4.132 4.132 0 0 0 4.149-4.148 4.132 4.132 0 0 0-4.149-4.148A4.13 4.13 0 0 0 0 37.639a4.13 4.13 0 0 0 4.148 4.148ZM21.97 58.379a4.13 4.13 0 0 0 4.148-4.148 4.13 4.13 0 0 0-4.148-4.148 4.132 4.132 0 0 0-4.149 4.148 4.132 4.132 0 0 0 4.149 4.148Z" fill="#fbfbfb" fill-rule="nonzero"></path> </svg> <span>Inhalt</span> </div> <svg class="article-menu__content__bg-pattern" height="100%" width="100%"> <filter class="filter-roughpaper" height="100%" id="roughpaper-34146" width="100%" x="0%" y="0%"> <feturbulence basefrequency="0.004" result="noise" seed="10"></feturbulence> <fediffuselighting in="noise" lighting-color="white" surfacescale="3.5"> <fedistantlight azimuth="140" elevation="8"></fedistantlight> </fediffuselighting> </filter> <rect fill="none" filter="url(#roughpaper-34146)" height="100%" width="100%" x="0" y="0"></rect> </svg> <div class="article-menu__content web-share flex justify-center items-center"> <a class="article-menu__content__link" href="https://the-decoder.de/diffsensei-forscher-entwickeln-ki-system-zur-automatischen-manga-generierung/#summary"> <img alt="summary" data-no-lazy="1" height="24" src="https://the-decoder.de/resources/icons/summary.svg" width="27"/> <span>Zusammenfassung</span> </a> </div> </div> </div> </div> <div class="entry-content__content-side"></div> <p><strong>Ein Forscherteam hat ein KI-System vorgestellt, das automatisch Manga-Comics erstellen kann. Das System namens DiffSensei kann Charaktere und Layouts präzise steuern und passt die Figuren dynamisch an die Geschichte an.</strong></p><div class="mobile-view"><div class="ad-notice">Anzeige</div> <div class="ad-row"> <div class="ad-container ad-m ad-feed" style="min-height: 600px;"> <div class="ad-m ad-feed" id="DEC_M_Incontent-1"></div> </div> </div> </div> <p>Wissenschaftler der Peking University, des Shanghai AI Laboratory und der Nanyang Technological University haben ein <a class="mixed-keyword" href="https://the-decoder.de/kuenstliche-intelligenz-begriffe-erklaerung/">KI</a>-System entwickelt, das automatisch Manga-Comics generieren kann. Dafür kombiniert das System "DiffSensei" Diffusions-Modelle mit großen Sprachmodellen, um Charaktere und Layouts präzise zu steuern.</p> <p>Die Forscher demonstrieren die Fähigkeiten des Systems auch mit einer fiktiven Geschichte über die KI-Pioniere Geoffrey Hinton, Yann LeCun und Yoshua Bengio. Die Geschichte erzählt, wie die drei Wissenschaftler versuchen, ein KI-Modell zu entwickeln, das die Transformer-Architektur übertrifft. Nach vielen Rückschlägen und Selbstzweifeln gelingt ihnen der Durchbruch - Jahre später werden sie dafür mit dem Nobelpreis ausgezeichnet.</p> <div class="content-img-side flex items-center gap-3"><figure aria-describedby="caption-attachment-34181" class="wp-caption alignnone" id="attachment_34181" style="width: 770px"><a href="https://the-decoder.de/wp-content/uploads/2024/12/DiffSensei-Example-1.png"><img alt="" class="wp-image-34181 size-medium" data-lazyloaded="1" data-src="https://the-decoder.de/wp-content/uploads/2024/12/DiffSensei-Example-1-770x242.png" decoding="async" fetchpriority="high" height="242" src="data:image/svg+xml;base64,PHN2ZyB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciIHdpZHRoPSI3NzAiIGhlaWdodD0iMjQyIiB2aWV3Qm94PSIwIDAgNzcwIDI0MiI+PHJlY3Qgd2lkdGg9IjEwMCUiIGhlaWdodD0iMTAwJSIgc3R5bGU9ImZpbGw6I2YyZjJmMjtmaWxsLW9wYWNpdHk6IDAuMTsiLz48L3N2Zz4=" width="770"/></a><figcaption class="wp-caption-text" id="caption-attachment-34181">Bild: Wu et al.</figcaption></figure> <div class="share flex items-start gap-2"> <svg class="share__icon switch" clip-rule="evenodd" fill-rule="evenodd" stroke-linejoin="round" stroke-miterlimit="2" viewbox="0 0 167 144" width="40" xml:space="preserve" xmlns="http://www.w3.org/2000/svg"> <path d="m75.517 1.317 58.028 41.334c1.59 1.059 2.385 2.915 2.385 4.769 0 1.855-.795 3.709-2.385 4.768L75.517 94.055c-1.855 1.325-4.24 1.59-6.093.531-1.856-1.059-3.179-3.179-3.179-5.299V65.705c0-3.18 2.65-5.83 5.829-5.83s5.829 2.651 5.829 5.83v11.923l41.865-29.942-41.867-29.942v11.923c0 3.179-2.384 5.829-5.565 5.829-40.804 2.12-54.583 23.052-58.823 45.044 9.009-10.069 19.342-16.163 31.796-18.813 3.179-.795 6.359 1.325 7.154 4.504.795 3.179-1.325 6.36-4.504 7.155-10.864 2.384-23.582 7.683-36.83 31.265-1.059 1.855-3.179 2.915-5.035 2.915-.531 0-1.059 0-1.59-.265-2.65-.795-4.504-2.914-4.504-5.829 0-12.19 1.059-30.47 10.334-46.369C20.932 36.819 40.01 26.222 66.242 23.571V5.554c0-2.12 1.325-4.24 3.179-5.299 1.856-.528 4.24-.264 6.096 1.061v.001Zm91.149 136.723V48.481c0-3.179-2.65-5.829-5.829-5.829h-8.742c-3.179 0-5.83 2.65-5.83 5.829s2.651 5.829 5.83 5.829h2.65v77.636H55.647V90.611c0-3.179-2.65-5.829-5.829-5.829s-5.83 2.651-5.83 5.829v47.43c0 3.179 2.651 5.829 5.83 5.829h111.021c3.179.002 5.828-2.649 5.828-5.827l-.001-.003Z" fill="#28293d" fill-rule="nonzero"></path> </svg> <div> <div class="share__title">Artikel teilen</div> <div class="share__info">Empfiehl unseren Artikel weiter</div> <button class="web-share share__button button button--filled">Teilen</button> </div> </div></div> <figure aria-describedby="caption-attachment-34182" class="wp-caption alignnone" id="attachment_34182" style="width: 770px"><a href="https://the-decoder.de/wp-content/uploads/2024/12/DiffSensei-Example-2.png"><img alt="" class="wp-image-34182 size-medium" data-lazyloaded="1" data-src="https://the-decoder.de/wp-content/uploads/2024/12/DiffSensei-Example-2-770x244.png" decoding="async" height="244" src="data:image/svg+xml;base64,PHN2ZyB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciIHdpZHRoPSI3NzAiIGhlaWdodD0iMjQ0IiB2aWV3Qm94PSIwIDAgNzcwIDI0NCI+PHJlY3Qgd2lkdGg9IjEwMCUiIGhlaWdodD0iMTAwJSIgc3R5bGU9ImZpbGw6I2YyZjJmMjtmaWxsLW9wYWNpdHk6IDAuMTsiLz48L3N2Zz4=" width="770"/></a><figcaption class="wp-caption-text" id="caption-attachment-34182">Bild: Wu et al.</figcaption></figure> <h2>DiffSensei generiert personalisierte Mangas</h2> <p>DiffSensei nutzt multimodale Modelle und Methoden wie LoRAs, um Charaktere identisch in einzelnen Bildern zu halten. Das Team generiert dann in einem dreistufigen Prozess Layouts, Zeichnungen mit festgelegten Charakteren und Dialoge.</p><div class="mobile-view"><div class="ad-notice">Anzeige</div> <div class="ad-row"> <div class="ad-container ad-m ad-feed" style="min-height: 600px;"> <div class="ad-m ad-feed" id="DEC_M_Incontent-2"></div> </div> </div> </div><div class="desktop-view"><div class="ad-notice">Anzeige</div> <div class="ad-row"> <div class="ad-container ad ad-feed" style="min-height: 280px;"> <div class="ad ad-feed" id="DEC_D_Incontent-1"></div> </div> <div class="ad-feed-side"> <div class="newsletter newsletter--feed flex gap-2"> <svg class="newsletter__svg switch" clip-rule="evenodd" fill-rule="evenodd" height="38" stroke-linejoin="round" stroke-miterlimit="2" viewbox="0 0 167 160" width="40" xml:space="preserve" xmlns="http://www.w3.org/2000/svg"> <path d="M103.175 86.95c-5.101-5.101-12.187-7.937-19.558-7.937-7.37 0-14.173 2.835-19.559 7.937l-6.519 6.518-47.052-31.745c-1.701-1.133-3.686-1.418-5.669-.85-1.985.567-3.4 1.985-3.968 3.968C.283 66.258 0 67.676 0 69.092v77.948c0 3.4 1.418 6.518 3.685 8.786 1.133 1.133 2.836 1.985 4.536 1.985 1.7 0 3.118-.567 4.536-1.985 0 0 0-.283.282-.283l59.808-59.807c2.835-2.835 6.518-4.251 10.487-4.251 3.968 0 7.653 1.418 10.487 4.251l51.02 51.02H37.131c-3.4 0-6.235 2.836-6.235 6.236 0 3.401 2.835 6.236 6.235 6.236h117.064c3.401 0 6.519-1.418 8.786-3.685 2.268-2.268 3.686-5.386 3.686-8.787l-.002-77.664c0-1.985-.568-4.251-1.418-5.954-.85-1.417-1.7-2.55-3.118-3.685L98.07 5.315c-8.504-7.087-20.408-7.087-28.912 0L20.972 45.849c-2.55 2.267-3.118 6.235-.85 9.071 2.268 2.55 6.236 3.118 9.072.85l48.185-40.532c3.685-3.118 9.071-3.118 12.472 0l53.855 45.068 6.803 5.668L125 83.264c-2.835 1.985-3.685 5.954-1.7 8.787 1.133 1.7 3.118 2.835 5.386 2.835 1.133 0 2.55-.283 3.685-1.133L154.48 78.73v59.524L103.175 86.95ZM48.47 102.54l-35.431 35.431V78.73l35.431 23.81Z" fill="#28293d" fill-rule="nonzero"></path> </svg> <div> <div class="newsletter__title">THE DECODER Newsletter</div> <div class="newsletter__description">Die wichtigen KI-News direkt ins E-Mail-Postfach.</div> <div>✓ 1x wöchentlich</div> <div>✓ kostenlos</div> <div>✓ jederzeit kündbar</div> <div class="newsletter__form"> <div class="mailpoet_form_popup_overlay"></div> <div class="mailpoet_form mailpoet_form_php mailpoet_form_position_ mailpoet_form_animation_" id="mailpoet_form_1"> <form action="https://the-decoder.de/wp-admin/admin-post.php?action=mailpoet_subscription_form" class="mailpoet_form mailpoet_form_form mailpoet_form_php" data-cookie-expiration-time="" data-delay="" data-exit-intent-enabled="" data-font-family="" method="post" novalidate="" target="_self"> <input name="data[form_id]" type="hidden" value="1"/> <input name="token" type="hidden" value="bca76b486f"/> <input name="api_version" type="hidden" value="v1"/> <input name="endpoint" type="hidden" value="subscribers"/> <input name="mailpoet_method" type="hidden" value="subscribe"/> <label class="mailpoet_hp_email_label" style="display: none !important;">Bitte dieses Feld leer lassen<input name="data[email]" type="email"/></label><div class="mailpoet_paragraph"><input aria-label="E-Mail-Adresse *" aria-required="true" autocomplete="email" class="mailpoet_text" data-automation-id="form_email" data-parsley-errors-container=".mailpoet_error_bevwr" data-parsley-maxlength="150" data-parsley-minlength="6" data-parsley-required="true" data-parsley-required-message="Dieses Feld wird benötigt." data-parsley-type-message="Dieser Wert sollte eine gültige E-Mail-Adresse sein." id="form_email_1" name="data[form_field_ODU0NzQzYzU3NmZmX2VtYWls]" placeholder="E-Mail-Adresse *" required="" style="width:100%;box-sizing:border-box;padding:5px;margin: 0 auto 0 0;" title="E-Mail-Adresse" type="email" value=""/><span class="mailpoet_error_bevwr"></span></div> <div class="mailpoet_paragraph"><input class="mailpoet_submit" data-automation-id="subscribe-submit-button" style="width:100%;box-sizing:border-box;padding:5px;margin: 0 auto 0 0;border-color:transparent;" type="submit" value="Kostenlos abonnieren"/><span class="mailpoet_form_loading"><span class="mailpoet_bounce1"></span><span class="mailpoet_bounce2"></span><span class="mailpoet_bounce3"></span></span></div> <div class="mailpoet_message"> <p class="mailpoet_validate_success" style="display:none;">Prüfen Sie Ihren Posteingang oder Spam-Ordner, um Ihr Abonnement zu bestätigen. </p> <p class="mailpoet_validate_error" style="display:none;"> </p> </div> </form> </div> </div> </div> </div> </div> </div> </div> <figure aria-describedby="caption-attachment-34183" class="wp-caption alignnone" id="attachment_34183" style="width: 770px"><img alt="" class="wp-image-34183 size-medium" data-lazyloaded="1" data-src="https://the-decoder.de/wp-content/uploads/2024/12/DiffSensei-method-770x421.png" decoding="async" height="421" src="data:image/svg+xml;base64,PHN2ZyB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciIHdpZHRoPSI3NzAiIGhlaWdodD0iNDIxIiB2aWV3Qm94PSIwIDAgNzcwIDQyMSI+PHJlY3Qgd2lkdGg9IjEwMCUiIGhlaWdodD0iMTAwJSIgc3R5bGU9ImZpbGw6I2YyZjJmMjtmaWxsLW9wYWNpdHk6IDAuMTsiLz48L3N2Zz4=" width="770"/><figcaption class="wp-caption-text" id="caption-attachment-34183">Bild: Wu et al.</figcaption></figure> <p>Für das Training des Systems haben die Forscher zudem einen neuen Datensatz namens MangaZero erstellt. Dieser umfasst 43.264 Manga-Seiten und 427.147 einzelne Panels aus 48 verschiedenen Manga-Serien.</p> <p>Die Panels wurden mit detaillierten Annotationen versehen, die unter anderem die Position der Charaktere und Dialoge markieren. Diese Informationen sind laut dem Team entscheidend für das Training des Systems.</p> <h2>Forscher sehen Potenzial in der Manga-Produktion</h2> <p>Die Forscher räumen in ihrer Studie auch Schwächen des Systems ein. Bei undeutlichen Eingabebildern der Charaktere hat das System Probleme, deren Erscheinungsbild korrekt zu erfassen. Außerdem kommt es bei mehreren ähnlich aussehenden Figuren manchmal zu einer unerwünschten "Verschmelzung" der Charaktere.</p> <p>Ohne Charakter-Eingabebilder hat das System zudem Schwierigkeiten, den Manga-Stil präzise zu kontrollieren. Die generierten Bilder tendieren dann zu einem eher generischen Manga-Look.</p> <p>Das System könnte laut den Forschern dennoch in Zukunft die Manga-Produktion unterstützen und den Arbeitsablauf optimieren. Die Technologie ermögliche es Künstlern, Verlagen und Kreativen, personalisierte Manga-Geschichten mit detaillierter Kontrolle über Charaktere und Layouts zu erstellen.</p><div class="article-recommendation-wrapper"><div class="article-recommendation">Empfehlung</div> <div class="card"> <div class="card__bg card__bg--tablet-featured gradient-transition"> <div class="card__bg__pattern-wrapper"> <svg class="card__bg__pattern" height="100%" width="100%"> <filter class="filter-roughpaper" height="100%" id="roughpaper-28288-3111" width="100%" x="0%" y="0%"> <feturbulence basefrequency="0.004" result="noise" seed="10"></feturbulence> <fediffuselighting in="noise" lighting-color="white" surfacescale="3.5"> <fedistantlight azimuth="140" elevation="8"></fedistantlight> </fediffuselighting> </filter> <rect fill="none" filter="url(#roughpaper-28288-3111)" height="100%" width="100%" x="0" y="0"></rect> </svg> </div> </div> <div class="card__content flex flex-col card__content-no-slide"> <a aria-label="KI-Studie erklärt Schwächen bei Schlussfolgerungen - und zeigt Lösungsansatz" class="link-overlay" href="https://the-decoder.de/ki-studie-erklaert-schwaechen-bei-schlussfolgerungen-und-zeigt-loesungsansatz/"></a> <div class="card__content__header flex flex-wrap items-center"> <div class="post-info-wrapper flex flex-wrap items-center gap-2"> <a class="card__content__button button button--outline" href="https://the-decoder.de/kuenstliche-intelligenz-news/ki-forschung/" rel="category tag">KI-Forschung</a> </div> </div> <h2 class="card__content__title"><a class="card__content__link" href="https://the-decoder.de/ki-studie-erklaert-schwaechen-bei-schlussfolgerungen-und-zeigt-loesungsansatz/" rel="bookmark">KI-Studie erklärt Schwächen bei Schlussfolgerungen - und zeigt Lösungsansatz</a></h2> <div class="card__content__img card__content__img--desk-none"> <a aria-hidden="true" class="post-thumbnail" href="https://the-decoder.de/ki-studie-erklaert-schwaechen-bei-schlussfolgerungen-und-zeigt-loesungsansatz/" tabindex="-1"> <img alt="KI-Studie erklärt Schwächen bei Schlussfolgerungen - und zeigt Lösungsansatz" data-lazyloaded="1" data-src="https://the-decoder.de/wp-content/uploads/2024/06/Grokked-Transformer-title-375x210.png" height="210" loading="lazy" src="data:image/svg+xml;base64,PHN2ZyB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciIHdpZHRoPSIzNzUiIGhlaWdodD0iMjEwIiB2aWV3Qm94PSIwIDAgMzc1IDIxMCI+PHJlY3Qgd2lkdGg9IjEwMCUiIGhlaWdodD0iMTAwJSIgc3R5bGU9ImZpbGw6I2YyZjJmMjtmaWxsLW9wYWNpdHk6IDAuMTsiLz48L3N2Zz4=" width="375"/> </a> </div> </div> <div class="card__side"> <div class="card__side__content"> <a aria-hidden="true" class="post-thumbnail" href="https://the-decoder.de/ki-studie-erklaert-schwaechen-bei-schlussfolgerungen-und-zeigt-loesungsansatz/" tabindex="-1"> <img alt="KI-Studie erklärt Schwächen bei Schlussfolgerungen - und zeigt Lösungsansatz" data-lazyloaded="1" data-src="https://the-decoder.de/wp-content/uploads/2024/06/Grokked-Transformer-title-375x210.png" height="210" loading="lazy" src="data:image/svg+xml;base64,PHN2ZyB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciIHdpZHRoPSIzNzUiIGhlaWdodD0iMjEwIiB2aWV3Qm94PSIwIDAgMzc1IDIxMCI+PHJlY3Qgd2lkdGg9IjEwMCUiIGhlaWdodD0iMTAwJSIgc3R5bGU9ImZpbGw6I2YyZjJmMjtmaWxsLW9wYWNpdHk6IDAuMTsiLz48L3N2Zz4=" width="375"/> </a> </div> </div> </div></div> <p>Mehr Beispiele und den Datensatz gibt es auf der <a href="https://jianzongwu.github.io/projects/diffsensei/" rel="noopener" target="_blank">DiffSensei-Projektseite.</a></p><div class="mobile-view"><div class="ad-notice">Anzeige</div> <div class="ad-row"> <div class="ad-container ad-m ad-feed" style="min-height: 600px;"> <div class="ad-m ad-feed" id="DEC_M_Incontent-3"></div> </div> </div> </div><div class="desktop-view"><div class="ad-notice">Anzeige</div> <div class="ad-row"> <div class="ad-container ad ad-feed" style="min-height: 280px;"> <div class="ad ad-feed" id="DEC_D_Incontent-2"></div> </div> <div class="ad-feed-side"> <div class="community flex gap-2"> <svg class="community__svg switch" clip-rule="evenodd" fill-rule="evenodd" height="35.44" stroke-linejoin="round" stroke-miterlimit="2" viewbox="0 0 167 145" width="40" xml:space="preserve" xmlns="http://www.w3.org/2000/svg"> <path d="M59.427 135.123c.81-3.166 3.781-5.277 7.293-4.486 5.402 1.055 11.076 1.583 16.748 1.583 39.167 0 71.042-26.919 71.042-60.171 0-33.253-31.875-60.172-71.042-60.172S12.425 38.796 12.425 72.049c0 13.724 5.402 26.654 15.667 37.475 1.621 1.583 1.892 3.958 1.351 6.069-1.62 4.487-4.051 8.181-6.753 11.613 4.861-.263 10.804-1.583 16.209-4.749 2.971-1.584 6.483-.792 8.373 2.111 1.62 2.903.81 6.334-2.161 8.181-12.696 7.389-26.472 6.861-33.496 6.069-3.782-.791-6.482-3.166-7.292-6.334-.81-3.166.541-6.598 3.512-8.446 3.512-2.375 6.484-5.543 8.373-9.501C5.674 102.133 0 87.354 0 72.048.269 32.463 37.547 0 83.468 0c45.922 0 83.199 32.461 83.199 72.048s-37.278 72.047-83.199 72.047c-6.483 0-12.965-.792-19.449-1.848-3.24-.79-5.132-3.958-4.591-7.124h-.001Zm58.346-79.965c3.241 0 5.943-2.639 5.943-5.806 0-3.166-2.702-5.806-5.943-5.806H99.675c-3.241 0-5.943 2.64-5.943 5.806 0 3.167 2.702 5.806 5.943 5.806h18.098ZM78.065 43.283H49.161c-3.241 0-5.943 2.64-5.943 5.806s2.702 5.806 5.943 5.806h28.904c3.241 0 5.943-2.64 5.943-5.806 0-3.168-2.702-5.806-5.943-5.806Zm2.431 28.766c0 3.166 2.702 5.806 5.943 5.806h31.335c3.24 0 5.942-2.64 5.942-5.806s-2.702-5.806-5.942-5.806H86.439c-3.241 0-5.943 2.64-5.943 5.806Zm-31.335 5.806h15.667c3.241 0 5.943-2.64 5.943-5.806s-2.702-5.806-5.943-5.806H49.161c-3.241 0-5.943 2.64-5.943 5.806s2.433 5.806 5.943 5.806Zm68.612 11.084h-10.266c-3.241 0-5.942 2.64-5.942 5.806s2.702 5.806 5.942 5.806h10.266c3.24 0 5.942-2.64 5.942-5.806.27-3.166-2.432-5.806-5.942-5.806Zm-68.612 11.876h36.737c3.241 0 5.943-2.64 5.943-5.807 0-3.166-2.702-5.806-5.943-5.806H49.161c-3.241 0-5.943 2.64-5.943 5.806-.269 3.169 2.433 5.807 5.943 5.807Z" fill="#28293d" fill-rule="nonzero"></path> </svg> <div> <div class="community__title">Community beitreten</div> <div class="community__text">Kommt in die DECODER-Community bei Discord,Reddit, Twitter und Co. - wir freuen uns auf euch!</div> <div class="social-icons flex items-center gap-1"> <a href="https://discord.gg/8VKkHAacn8" rel="noopener noreferrer" target="_blank" title="Discord"> <svg class="switch" clip-rule="evenodd" fill-rule="evenodd" height="30" stroke-linejoin="round" stroke-miterlimit="2" viewbox="0 0 125 125" width="30" xml:space="preserve" xmlns="http://www.w3.org/2000/svg"> <path d="M62.5 0C96.995 0 125 28.005 125 62.5c0 34.495-28.005 62.5-62.5 62.5C28.005 125 0 96.995 0 62.5 0 28.005 28.005 0 62.5 0Zm30.322 34.681a72.201 72.201 0 0 0-17.81-5.524.272.272 0 0 0-.286.136c-.769 1.368-1.621 3.152-2.218 4.555-6.725-1.007-13.416-1.007-20.004 0-.596-1.434-1.479-3.187-2.252-4.555a.281.281 0 0 0-.286-.136 72.005 72.005 0 0 0-17.811 5.524.26.26 0 0 0-.117.101C20.695 51.729 17.587 68.26 19.112 84.586a.295.295 0 0 0 .113.204c7.485 5.497 14.734 8.833 21.849 11.045a.282.282 0 0 0 .307-.101 51.745 51.745 0 0 0 4.47-7.27.277.277 0 0 0-.151-.386 47.699 47.699 0 0 1-6.826-3.253.28.28 0 0 1-.028-.465 37.58 37.58 0 0 0 1.356-1.063.27.27 0 0 1 .283-.038c14.32 6.538 29.823 6.538 43.974 0a.27.27 0 0 1 .286.035c.438.361.897.722 1.359 1.066a.28.28 0 0 1-.024.465 44.852 44.852 0 0 1-6.829 3.25.28.28 0 0 0-.148.389 58.267 58.267 0 0 0 4.466 7.267.277.277 0 0 0 .307.104c7.149-2.212 14.399-5.548 21.883-11.045a.278.278 0 0 0 .114-.201c1.825-18.874-3.056-35.269-12.937-49.803a.22.22 0 0 0-.114-.105ZM47.99 74.645c-4.312 0-7.864-3.958-7.864-8.819 0-4.861 3.483-8.819 7.864-8.819 4.414 0 7.932 3.993 7.863 8.819 0 4.861-3.483 8.819-7.863 8.819Zm29.074 0c-4.311 0-7.863-3.958-7.863-8.819 0-4.861 3.483-8.819 7.863-8.819 4.415 0 7.933 3.993 7.864 8.819 0 4.861-3.449 8.819-7.864 8.819Z" fill="#28293d"></path> </svg> </a> <a href="https://twitter.com/TheDecoderDE" rel="noopener noreferrer" target="_blank" title="Twitter"> <svg class="switch" clip-rule="evenodd" fill-rule="evenodd" height="30" stroke-linejoin="round" stroke-miterlimit="2" viewbox="0 0 125 125" width="30" xml:space="preserve" xmlns="http://www.w3.org/2000/svg"> <path d="M125 62.5c0 34.518-27.982 62.5-62.501 62.5C27.981 125 0 97.018 0 62.5S27.981 0 62.499 0C97.018 0 125 27.982 125 62.5ZM49.869 98.495c30.529 0 47.224-25.289 47.224-47.219 0-.719-.019-1.434-.047-2.145a33.766 33.766 0 0 0 8.278-8.593 33.154 33.154 0 0 1-9.528 2.612 16.655 16.655 0 0 0 7.297-9.181 33.3 33.3 0 0 1-10.537 4.029 16.599 16.599 0 0 0-12.121-5.243c-9.166 0-16.592 7.432-16.592 16.594 0 1.303.139 2.57.425 3.784-13.797-.693-26.018-7.297-34.203-17.339a16.543 16.543 0 0 0-2.25 8.341 16.568 16.568 0 0 0 7.389 13.815 16.46 16.46 0 0 1-7.519-2.077c-.009.07-.009.139-.009.212 0 8.038 5.722 14.748 13.315 16.271a16.547 16.547 0 0 1-7.491.284c2.111 6.593 8.241 11.392 15.5 11.527a33.284 33.284 0 0 1-20.611 7.103c-1.333 0-2.658-.076-3.954-.229a46.977 46.977 0 0 0 25.434 7.454Z" fill="#28293d"></path> </svg> </a> <a href="https://www.facebook.com/TheDecoderDE/" rel="noopener noreferrer" target="_blank" title="Facebook"> <svg class="switch" clip-rule="evenodd" fill-rule="evenodd" height="30" stroke-linejoin="round" stroke-miterlimit="2" viewbox="0 0 125 125" width="30" xml:space="preserve" xmlns="http://www.w3.org/2000/svg"> <path d="M52.735 124.24C22.871 119.546 0 93.673 0 62.5 0 28.005 28.005 0 62.5 0 96.995 0 125 28.005 125 62.5c0 31.173-22.871 57.046-52.735 61.74V80.571h14.564L89.597 62.5H72.265V50.775c0-4.939 2.417-9.765 10.186-9.765h7.884V25.629s-7.154-1.221-13.992-1.221c-14.274 0-23.608 8.648-23.608 24.319V62.5H36.862v18.071h15.873v43.669Z" fill="#28293d"></path> </svg> </a> <a href="https://www.reddit.com/r/TheDecoder/" rel="noopener noreferrer" target="_blank" title="Reddit"> <svg class="switch" clip-rule="evenodd" fill-rule="evenodd" height="30" stroke-linejoin="round" stroke-miterlimit="2" viewbox="0 0 125 125" width="30" xml:space="preserve" xmlns="http://www.w3.org/2000/svg"> <path d="M62.5 0C96.995 0 125 28.005 125 62.5c0 34.495-28.005 62.5-62.5 62.5C28.005 125 0 96.995 0 62.5 0 28.005 28.005 0 62.5 0Zm41.692 62.5c0-5.072-4.103-9.099-9.1-9.099-2.461 0-4.698.97-6.339 2.536-6.265-4.475-14.842-7.384-24.388-7.757l4.176-19.54 13.574 2.908a6.486 6.486 0 0 0 6.489 6.191 6.493 6.493 0 0 0 6.488-6.489 6.493 6.493 0 0 0-6.488-6.489 6.487 6.487 0 0 0-5.818 3.655l-15.14-3.207c-.447-.075-.895 0-1.193.224-.373.223-.597.596-.671 1.044l-4.624 21.778c-9.696.298-18.422 3.207-24.762 7.756a9.161 9.161 0 0 0-6.339-2.535c-5.072 0-9.099 4.102-9.099 9.099 0 3.729 2.237 6.861 5.37 8.278-.15.895-.224 1.79-.224 2.76 0 14.021 16.333 25.432 36.471 25.432 20.137 0 36.47-11.336 36.47-25.432 0-.895-.074-1.865-.223-2.76a9.187 9.187 0 0 0 5.37-8.353ZM78.013 86.217c-4.475 4.475-12.977 4.773-15.438 4.773-2.536 0-11.039-.372-15.439-4.773a1.666 1.666 0 0 1 0-2.386 1.665 1.665 0 0 1 2.387 0c2.834 2.834 8.8 3.803 13.052 3.803 4.251 0 10.292-.969 13.051-3.803a1.665 1.665 0 0 1 2.387 0 1.813 1.813 0 0 1 0 2.386ZM76.82 75.552a6.494 6.494 0 0 1-6.489-6.489 6.493 6.493 0 0 1 6.489-6.488 6.492 6.492 0 0 1 6.488 6.488c0 3.506-2.908 6.489-6.488 6.489Zm-35.128-6.563A6.493 6.493 0 0 1 48.18 62.5a6.494 6.494 0 0 1 6.489 6.489 6.493 6.493 0 0 1-6.489 6.488c-3.58.075-6.488-2.908-6.488-6.488Z" fill="#28293d"></path> </svg> </a> <a href="https://www.linkedin.com/company/the-decoder-de/" rel="noopener noreferrer" target="_blank" title="LinkedIn"> <svg class="switch" clip-rule="evenodd" fill-rule="evenodd" height="30" stroke-linejoin="round" stroke-miterlimit="2" viewbox="0 0 125 125" width="30" xml:space="preserve" xmlns="http://www.w3.org/2000/svg"> <path d="M62.5 0C96.995 0 125 28.005 125 62.5c0 34.495-28.005 62.5-62.5 62.5C28.005 125 0 96.995 0 62.5 0 28.005 28.005 0 62.5 0Zm39.931 98.958H86.735V72.225c0-7.329-2.785-11.425-8.586-11.425-6.311 0-9.609 4.262-9.609 11.425v26.733H53.414V48.032H68.54v6.86s4.548-8.415 15.355-8.415c10.802 0 18.536 6.596 18.536 20.238v32.243Zm-74.872 0h15.772V48.032H27.559v50.926Zm7.81-57.594c-5.152 0-9.327-4.208-9.327-9.397 0-5.19 4.175-9.398 9.327-9.398s9.325 4.208 9.325 9.398c0 5.189-4.173 9.397-9.325 9.397Z" fill="#28293d"></path> </svg> </a> </div> </div> </div> </div> </div> </div><div id="wp-worthy-pixel"><img alt="" class="wp-worthy-pixel-img skip-lazy" data-no-lazy="1" data-skip-lazy="1" decoding="async" height="1" loading="eager" src="https://vg02.met.vgwort.de/na/e3c3d4ac525940e6849b90af13e53e9b" width="1"/></div> <div class="video-block responsive-container"><iframe allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" allowfullscreen="" data-src="https://www.youtube.com/embed/TLJ0MYZmoXc?feature=oembed" frameborder="0" height="433" referrerpolicy="strict-origin-when-cross-origin" title="DiffSensei: Bridging Multi-Modal LLMs and Diffusion Models for Customized Manga Generation" width="770"></iframe></div> <div></div>

DiffSensei: Bridging Multi-Modal LLMs and Diffusion Models for Customized Manga Generation