arXiv:2510.22196v1 Announce Type: new 
Abstract: Scaling and architectural advances have produced strikingly photorealistic image generative models, yet their mechanisms still remain opaque. Rather than advancing scaling, our goal is to strip away complicated engineering tricks and propose a simple, non-parametric generative model. Our design is grounded in three principles of natural images-(i) spatial non-stationarity, (ii) low-level regularities, and (iii) high-level semantics-and defines each pixel's distribution from its local context window. Despite its minimal architecture and no training, the model produces high-fidelity samples on MNIST and visually compelling CIFAR-10 images. This combination of simplicity and strong empirical performance points toward a minimal theory of natural-image structure. The model's white-box nature also allows us to have a mechanistic understanding of how the model generalizes and generates diverse images. We study it by tracing each generated pixel back to its source images. These analyses reveal a simple, compositional procedure for "part-whole generalization", suggesting a hypothesis for how large neural network generative models learn to generalize.

ظهر نهج جديد في النمذجة التوليدية يركز على البساطة والشفافية بدلاً من التقنيات الهندسية المعقدة. يعتمد هذا النموذج على مبادئ رئيسية للصور الطبيعية، بهدف تعزيز فهمنا لتوليد الصور. من خلال إزالة التعقيدات غير الضرورية، يمكن أن تؤدي هذه الأبحاث إلى طرق أكثر سهولة وكفاءة لإنشاء صور فوتوغرافية واقعية، وهو ما يعد مهمًا لمجموعة متنوعة من التطبيقات في التكنولوجيا والفن.

Ha surgido un nuevo enfoque para la modelización generativa, centrado en la simplicidad y la transparencia en lugar de en técnicas de ingeniería complejas. Este modelo se basa en principios clave de las imágenes naturales, con el objetivo de mejorar nuestra comprensión de la generación de imágenes. Al eliminar complicaciones innecesarias, esta investigación podría conducir a métodos más accesibles y eficientes para crear imágenes fotorealistas, lo cual es significativo para diversas aplicaciones en tecnología y arte.

Une nouvelle approche de la modélisation générative a émergé, mettant l'accent sur la simplicité et la transparence plutôt que sur des techniques d'ingénierie complexes. Ce modèle est basé sur des principes clés des images naturelles, visant à améliorer notre compréhension de la génération d'images. En éliminant les complications inutiles, cette recherche pourrait conduire à des méthodes plus accessibles et efficaces pour créer des images photoréalistes, ce qui est significatif pour diverses applications dans la technologie et l'art.

A new approach to generative modeling has emerged, focusing on simplicity and transparency rather than complex engineering. This model is based on key principles of natural images, aiming to enhance our understanding of image generation. By stripping away unnecessary complications, this research could lead to more accessible and efficient methods for creating photorealistic images, which is significant for various applications in technology and art.

Scaling Non-Parametric Sampling with Representation

One More Thing in AI – Your Shortcut to AI Mastery

Scaling Non-Parametric Sampling with Representation

Was this article worth reading? Share it

One More Thing in AI

ScreenshotOne

BeautiAI

Blunge

FLUX AI ART

Bulk Image Generation AI

Ready to build your own newsroom?