arXiv:2511.11692v1 Announce Type: new 
Abstract: Optimization-based text-to-3D methods distill guidance from 2D generative models via Score Distillation Sampling (SDS), but implicitly treat this guidance as static. This work shows that ignoring source dynamics yields inconsistent trajectories that suppress or merge semantic cues, leading to "semantic over-smoothing" artifacts. As such, we reformulate text-to-3D optimization as mapping a dynamically evolving source distribution to a fixed target distribution. We cast the problem into a dual-conditioned latent space, conditioned on both the text prompt and the intermediately rendered image. Given this joint setup, we observe that the image condition naturally anchors the current source distribution. Building on this insight, we introduce AnchorDS, an improved score distillation mechanism that provides state-anchored guidance with image conditions and stabilizes generation. We further penalize erroneous source estimates and design a lightweight filter strategy and fine-tuning strategy that refines the anchor with negligible overhead. AnchorDS produces finer-grained detail, more natural colours, and stronger semantic consistency, particularly for complex prompts, while maintaining efficiency. Extensive experiments show that our method surpasses previous methods in both quality and efficiency.

AnchorDS هو نهج مبتكر لتوليد النص إلى 3D يعالج قيود الطرق الحالية المعتمدة على التحسين، والتي غالبًا ما تعتبر الإرشادات من النماذج التوليدية ثنائية الأبعاد ثابتة. تسلط هذه الدراسة الضوء على مشكلة 'تنعيم المعاني' الناتجة عن تجاهل ديناميات المصادر. من خلال إعادة صياغة عملية التحسين لتعيين توزيع مصدر يتطور ديناميكيًا إلى توزيع هدف ثابت، يقدم AnchorDS مساحة كامنة ذات شرط مزدوج تعمل على استقرار التوليد من خلال توجيه مثبت في الحالة.

AnchorDS es un enfoque novedoso para la generación de texto a 3D que aborda las limitaciones de los métodos de optimización existentes, que a menudo tratan la guía de modelos generativos 2D como estática. Esta investigación destaca el problema de los artefactos de 'suavizado semántico' causados por ignorar la dinámica de las fuentes. Al reformular el proceso de optimización para mapear una distribución de fuente en evolución dinámica a una distribución objetivo fija, AnchorDS introduce un espacio latente de doble condición que estabiliza la generación a través de una guía anclada en el estado.

AnchorDS est une approche novatrice pour la génération de texte en 3D qui aborde les limitations des méthodes d'optimisation existantes, souvent considérées comme statiques. Cette recherche met en lumière le problème des artefacts de 'lissage sémantique' causés par l'ignorance de la dynamique des sources. En reformulant le processus d'optimisation pour cartographier une distribution source en évolution dynamique vers une distribution cible fixe, AnchorDS introduit un espace latent à double condition qui stabilise la génération grâce à un guidage ancré dans l'état.

AnchorDS is a novel approach to text-to-3D generation that addresses the limitations of existing optimization-based methods, which often treat guidance from 2D generative models as static. This research highlights the issue of 'semantic over-smoothing' artifacts caused by ignoring source dynamics. By reformulating the optimization process to map a dynamically evolving source distribution to a fixed target distribution, AnchorDS introduces a dual-conditioned latent space that stabilizes generation through state-anchored guidance.

AnchorDS: Anchoring Dynamic Sources for Semantically Consistent Text-to-3D Generation

arXiv:2511.14899v1 Announce Type: new 
Abstract: We address the task of multi-view image editing from sparse input views, where the inputs can be seen as a mix of images capturing the scene from different viewpoints. The goal is to modify the scene according to a textual instruction while preserving consistency across all views. Existing methods, based on per-scene neural fields or temporal attention mechanisms, struggle in this setting, often producing artifacts and incoherent edits. We propose InstructMix2Mix (I-Mix2Mix), a framework that distills the editing capabilities of a 2D diffusion model into a pretrained multi-view diffusion model, leveraging its data-driven 3D prior for cross-view consistency. A key contribution is replacing the conventional neural field consolidator in Score Distillation Sampling (SDS) with a multi-view diffusion student, which requires novel adaptations: incremental student updates across timesteps, a specialized teacher noise scheduler to prevent degeneration, and an attention modification that enhances cross-view coherence without additional cost. Experiments demonstrate that I-Mix2Mix significantly improves multi-view consistency while maintaining high per-frame edit quality.

يقدم المقال InstructMix2Mix (I-Mix2Mix)، وهو إطار مصمم لتحرير الصور متعددة المناظر من مدخلات نادرة. يهدف إلى تعديل المشاهد وفقًا لتعليمات نصية مع الحفاظ على التناسق عبر وجهات نظر مختلفة. غالبًا ما تنتج الطرق الحالية عيوبًا وتعديلات غير متسقة، لكن I-Mix2Mix يستفيد من نموذج انتشار متعدد المناظر مدرب مسبقًا لتعزيز التناسق عبر المناظر، مما يعالج هذه التحديات بشكل فعال.

El artículo presenta InstructMix2Mix (I-Mix2Mix), un marco diseñado para la edición de imágenes desde vistas de entrada escasas. Su objetivo es modificar escenas según instrucciones textuales mientras se asegura la consistencia entre diferentes puntos de vista. Los métodos existentes a menudo producen artefactos y ediciones incoherentes, pero I-Mix2Mix aprovecha un modelo de difusión multi-vista preentrenado para mejorar la consistencia entre vistas, abordando estos desafíos de manera efectiva.

L'article présente InstructMix2Mix (I-Mix2Mix), un cadre conçu pour l'édition d'images multi-vues à partir de vues d'entrée rares. Il vise à modifier des scènes en fonction d'instructions textuelles tout en garantissant la cohérence entre différents points de vue. Les méthodes existantes produisent souvent des artefacts et des modifications incohérentes, mais I-Mix2Mix exploite un modèle de diffusion multi-vues pré-entraîné pour améliorer la cohérence entre les vues, répondant ainsi efficacement à ces défis.

The article presents InstructMix2Mix (I-Mix2Mix), a framework designed for multi-view image editing from sparse input views. It aims to modify scenes based on textual instructions while ensuring consistency across various viewpoints. Existing methods often produce artifacts and incoherent edits, but I-Mix2Mix leverages a pretrained multi-view diffusion model to enhance cross-view consistency, addressing these challenges effectively.

InstructMix2Mix: Consistent Sparse-View Editing Through Multi-View Model Personalization

arXiv:2511.01767v2 Announce Type: replace 
Abstract: In this work, we introduce \textbf{Wonder3D++}, a novel method for efficiently generating high-fidelity textured meshes from single-view images. Recent methods based on Score Distillation Sampling (SDS) have shown the potential to recover 3D geometry from 2D diffusion priors, but they typically suffer from time-consuming per-shape optimization and inconsistent geometry. In contrast, certain works directly produce 3D information via fast network inferences, but their results are often of low quality and lack geometric details. To holistically improve the quality, consistency, and efficiency of single-view reconstruction tasks, we propose a cross-domain diffusion model that generates multi-view normal maps and the corresponding color images. To ensure the consistency of generation, we employ a multi-view cross-domain attention mechanism that facilitates information exchange across views and modalities. Lastly, we introduce a cascaded 3D mesh extraction algorithm that drives high-quality surfaces from the multi-view 2D representations in only about $3$ minute in a coarse-to-fine manner. Our extensive evaluations demonstrate that our method achieves high-quality reconstruction results, robust generalization, and good efficiency compared to prior works. Code available at https://github.com/xxlong0/Wonder3D/tree/Wonder3D_Plus.

Wonder3D++ هي طريقة جديدة مصممة لتوليد شبكات عالية الجودة من الصور ذات العرض الواحد. تتناول القيود المفروضة على الأساليب الحالية التي تتطلب تحسينًا مكثفًا أو تنتج نتائج منخفضة الجودة. من خلال استخدام نموذج انتشار عبر المجالات وآلية انتباه متعددة الرؤى، تعمل Wonder3D++ على تحسين جودة وموثوقية إعادة بناء الأبعاد الثلاثية، مما يمثل تقدمًا كبيرًا في مجال توليد الأبعاد الثلاثية.

Wonder3D++ es un nuevo método diseñado para generar mallas texturizadas de alta fidelidad a partir de imágenes de vista única. Aborda las limitaciones de las técnicas existentes que requieren optimización extensa o producen resultados de baja calidad. Al emplear un modelo de difusión interdominio y un mecanismo de atención multivista, Wonder3D++ mejora la calidad y consistencia de las reconstrucciones 3D, representando un avance significativo en el campo de la generación 3D.

Wonder3D++ est une nouvelle méthode conçue pour générer des maillages texturés de haute fidélité à partir d'images à vue unique. Elle répond aux limitations des techniques existantes qui nécessitent soit une optimisation extensive, soit produisent des résultats de faible qualité. En utilisant un modèle de diffusion inter-domaines et un mécanisme d'attention multi-vues, Wonder3D++ améliore la qualité et la cohérence des reconstructions 3D, représentant une avancée significative dans le domaine de la génération 3D.

Wonder3D++ is a new method designed to generate high-fidelity textured meshes from single-view images. It addresses limitations in existing techniques that either require extensive optimization or yield low-quality results. By employing a cross-domain diffusion model and a multi-view attention mechanism, Wonder3D++ enhances the quality and consistency of 3D reconstructions, making it a significant advancement in the field of 3D generation.

AnchorDS: Anchoring Dynamic Sources for Semantically Consistent Text-to-3D Generation

Was this article worth reading? Share it