arXiv:2506.10036v2 Announce Type: replace-cross 
Abstract: Classifier-free guidance (CFG) has become an essential component of modern diffusion models to enhance both generation quality and alignment with input conditions. However, CFG requires specific training procedures and is limited to conditional generation. To address these limitations, we propose Token Perturbation Guidance (TPG), a novel method that applies perturbation matrices directly to intermediate token representations within the diffusion network. TPG employs a norm-preserving shuffling operation to provide effective and stable guidance signals that improve generation quality without architectural changes. As a result, TPG is training-free and agnostic to input conditions, making it readily applicable to both conditional and unconditional generation. We further analyze the guidance term provided by TPG and show that its effect on sampling more closely resembles CFG compared to existing training-free guidance techniques. Extensive experiments on SDXL and Stable Diffusion 2.1 show that TPG achieves nearly a 2$\times$ improvement in FID for unconditional generation over the SDXL baseline, while closely matching CFG in prompt alignment. These results establish TPG as a general, condition-agnostic guidance method that brings CFG-like benefits to a broader class of diffusion models.

تم تقديم طريقة جديدة تُدعى توجيه الاضطراب الرمزي (TPG) لتحسين نماذج الانتشار، التي تعتبر ضرورية لتوليد نتائج عالية الجودة. على عكس التوجيه الخالي من المصنفات (CFG) الموجود، الذي له قيود في التدريب والتوليد الشرطي، يقوم TPG بتطبيق مصفوفات الاضطراب مباشرة على تمثيلات الرموز، مما يعزز جودة ومواءمة المحتوى المُولد. هذه الابتكار مهم لأنه يفتح آفاقًا جديدة لتدريب النماذج بشكل أكثر فعالية ومرونة، مما قد يؤدي إلى تقدم في تطبيقات الذكاء الاصطناعي المختلفة.

Se ha introducido un nuevo método llamado Token Perturbation Guidance (TPG) para mejorar los modelos de difusión, que son cruciales para generar resultados de alta calidad. A diferencia de la Classifier-free Guidance (CFG) existente, que tiene limitaciones en el entrenamiento y la generación condicional, TPG aplica matrices de perturbación directamente a las representaciones de tokens, mejorando tanto la calidad como la alineación del contenido generado. Esta innovación es significativa ya que abre nuevas posibilidades para un entrenamiento de modelos más efectivo y flexible, lo que podría llevar a avances en diversas aplicaciones de IA.

Une nouvelle méthode appelée Token Perturbation Guidance (TPG) a été introduite pour améliorer les modèles de diffusion, qui sont essentiels pour générer des résultats de haute qualité. Contrairement à la Classifier-free Guidance (CFG) existante, qui présente des limitations en matière de formation et de génération conditionnelle, TPG applique des matrices de perturbation directement aux représentations de tokens, améliorant ainsi la qualité et l'alignement du contenu généré. Cette innovation est significative car elle ouvre de nouvelles possibilités pour une formation de modèle plus efficace et flexible, ce qui pourrait conduire à des avancées dans diverses applications de l'IA.

A new method called Token Perturbation Guidance (TPG) has been introduced to improve diffusion models, which are crucial for generating high-quality outputs. Unlike the existing Classifier-free Guidance (CFG) that has limitations in training and conditional generation, TPG applies perturbation matrices directly to token representations, enhancing both the quality and alignment of generated content. This innovation is significant as it opens up new possibilities for more effective and flexible model training, potentially leading to advancements in various applications of AI.

Token Perturbation Guidance for Diffusion Models

arXiv:2512.03540v1 Announce Type: new 
Abstract: Cooking is a sequential and visually grounded activity, where each step such as chopping, mixing, or frying carries both procedural logic and visual semantics. While recent diffusion models have shown strong capabilities in text-to-image generation, they struggle to handle structured multi-step scenarios like recipe illustration. Additionally, current recipe illustration methods are unable to adjust to the natural variability in recipe length, generating a fixed number of images regardless of the actual instructions structure. To address these limitations, we present CookAnything, a flexible and consistent diffusion-based framework that generates coherent, semantically distinct image sequences from textual cooking instructions of arbitrary length. The framework introduces three key components: (1) Step-wise Regional Control (SRC), which aligns textual steps with corresponding image regions within a single denoising process; (2) Flexible RoPE, a step-aware positional encoding mechanism that enhances both temporal coherence and spatial diversity; and (3) Cross-Step Consistency Control (CSCC), which maintains fine-grained ingredient consistency across steps. Experimental results on recipe illustration benchmarks show that CookAnything performs better than existing methods in training-based and training-free settings. The proposed framework supports scalable, high-quality visual synthesis of complex multi-step instructions and holds significant potential for broad applications in instructional media, and procedural content creation.

تم تقديم CookAnything كإطار جديد مصمم لتوليد تسلسلات صور متماسكة ومتميزة دلاليًا من التعليمات النصية للطهي بأطوال متنوعة، مما يعالج قيود نماذج الانتشار الحالية في توضيح الوصفات متعددة الخطوات.

CookAnything se ha presentado como un nuevo marco diseñado para generar secuencias de imágenes coherentes y semánticamente distintas a partir de instrucciones de cocina textuales de diversas longitudes, abordando las limitaciones de los modelos de difusión actuales en la ilustración de recetas de múltiples pasos.

CookAnything a été introduit comme un nouveau cadre conçu pour générer des séquences d'images cohérentes et sémantiquement distinctes à partir d'instructions de cuisine textuelles de longueurs variées, répondant aux limitations des modèles de diffusion actuels dans l'illustration de recettes multi-étapes.

CookAnything has been introduced as a novel framework designed to generate coherent and semantically distinct image sequences from textual cooking instructions of varying lengths, addressing the limitations of current diffusion models in multi-step recipe illustration.

CookAnything: A Framework for Flexible and Consistent Multi-Step Recipe Image Generation

arXiv:2512.03451v1 Announce Type: cross 
Abstract: Diffusion models have revolutionized video generation, becoming essential tools in creative content generation and physical simulation. Transformer-based architectures (DiTs) and classifier-free guidance (CFG) are two cornerstones of this success, enabling strong prompt adherence and realistic video quality. Despite their versatility and superior performance, these models require intensive computation. Each video generation requires dozens of iterative steps, and CFG doubles the required compute. This inefficiency hinders broader adoption in downstream applications.
  We introduce GalaxyDiT, a training-free method to accelerate video generation with guidance alignment and systematic proxy selection for reuse metrics. Through rank-order correlation analysis, our technique identifies the optimal proxy for each video model, across model families and parameter scales, thereby ensuring optimal computational reuse. We achieve $1.87\times$ and $2.37\times$ speedup on Wan2.1-1.3B and Wan2.1-14B with only 0.97% and 0.72% drops on the VBench-2.0 benchmark. At high speedup rates, our approach maintains superior fidelity to the base model, exceeding prior state-of-the-art approaches by 5 to 10 dB in peak signal-to-noise ratio (PSNR).

تم تقديم GalaxyDiT كطريقة خالية من التدريب لتحسين كفاءة توليد الفيديو باستخدام محولات الانتشار، حيث تعالج عدم الكفاءة الحاسوبية المرتبطة بالنماذج الحالية التي تتطلب خطوات تكرارية عديدة وموارد. تركز هذه الابتكار على محاذاة الإرشادات واختيار الوكيل التكيفي لتحسين إعادة استخدام الحوسبة عبر عائلات النماذج المختلفة.

Se ha presentado GalaxyDiT como un método sin entrenamiento para mejorar la eficiencia de la generación de video utilizando transformadores de difusión, abordando las ineficiencias computacionales asociadas con los modelos existentes que requieren numerosos pasos iterativos y recursos. Esta innovación se centra en la alineación de la guía y la selección adaptativa de proxy para optimizar la reutilización computacional entre diferentes familias de modelos.

GalaxyDiT a été présenté comme une méthode sans entraînement pour améliorer l'efficacité de la génération vidéo utilisant des transformateurs de diffusion, abordant les inefficacités computationnelles associées aux modèles existants qui nécessitent de nombreuses étapes itératives et ressources. Cette innovation se concentre sur l'alignement des directives et la sélection de proxy adaptative pour optimiser la réutilisation computationnelle à travers différentes familles de modèles.

GalaxyDiT has been introduced as a training-free method to enhance the efficiency of video generation using diffusion transformers, addressing the computational inefficiencies associated with existing models that require extensive iterative steps and resources. This innovation focuses on guidance alignment and adaptive proxy selection to optimize computational reuse across different model families.

Token Perturbation Guidance for Diffusion Models

Was this article worth reading? Share it

LucidQuery AI

Airparser

GPTHumanizer