arXiv:2511.21185v1 Announce Type: new 
Abstract: Recent visual autoregressive (AR) models have shown promising capabilities in text-to-image generation, operating in a manner similar to large language models. While test-time computation scaling has brought remarkable success in enabling reasoning-enhanced outputs for challenging natural language tasks, its adaptation to visual AR models remains unexplored and poses unique challenges. Naively applying test-time scaling strategies such as Best-of-N can be suboptimal: they consume full-length computation on erroneous generation trajectories, while the raster-scan decoding scheme lacks a blueprint of the entire canvas, limiting scaling benefits as only a few prompt-aligned candidates are generated. To address these, we introduce GridAR, a test-time scaling framework designed to elicit the best possible results from visual AR models. GridAR employs a grid-partitioned progressive generation scheme in which multiple partial candidates for the same position are generated within a canvas, infeasible ones are pruned early, and viable ones are fixed as anchors to guide subsequent decoding. Coupled with this, we present a layout-specified prompt reformulation strategy that inspects partial views to infer a feasible layout for satisfying the prompt. The reformulated prompt then guides subsequent image generation to mitigate the blueprint deficiency. Together, GridAR achieves higher-quality results under limited test-time scaling: with N=4, it even outperforms Best-of-N (N=8) by 14.4% on T2I-CompBench++ while reducing cost by 25.6%. It also generalizes to autoregressive image editing, showing comparable edit quality and a 13.9% gain in semantic preservation on PIE-Bench over larger-N baselines.

أدت التطورات الأخيرة في النماذج التلقائية الانحدار المرئية إلى تقديم GridAR، وهو إطار عمل لتوسيع نطاق الوقت الاختباري يهدف إلى تحسين توليد الصور من النص. يتناول هذا الإطار القيود المفروضة على الاستراتيجيات الحالية من خلال استخدام مخطط توليد تدريجي مقسم إلى شبكة، مما يسمح بتحسين النتائج المرشحة خلال عملية التوليد.

Los avances recientes en modelos autoregresivos visuales han llevado a la introducción de GridAR, un marco de escalado en tiempo de prueba diseñado para mejorar la generación de imágenes a partir de texto. Este marco aborda las limitaciones de las estrategias existentes mediante un esquema de generación progresiva particionado en cuadrícula, lo que permite mejorar las salidas candidatas durante el proceso de generación.

Des avancées récentes dans les modèles autoregressifs visuels ont conduit à l'introduction de GridAR, un cadre de mise à l'échelle au moment du test visant à améliorer la génération d'images à partir de texte. Ce cadre aborde les limitations des stratégies existantes en employant un schéma de génération progressive partitionné en grille, permettant d'améliorer les sorties candidates lors du processus de génération.

Recent advancements in visual autoregressive models have led to the introduction of GridAR, a test-time scaling framework aimed at enhancing text-to-image generation. This framework addresses the limitations of existing strategies by employing a grid-partitioned progressive generation scheme, allowing for improved candidate outputs during the generation process.

Progress by Pieces: Test-Time Scaling for Autoregressive Image Generation

Was this article worth reading? Share it

Tattoo Visualizer

PicPicAI

FETCH HIVE