Adaptive Visual Autoregressive Acceleration via Dual-Linkage Entropy Analysis

Visual AutoRegressive modeling (VAR) suffers from substantial computational cost due to the massive token count involved. Failing to account for the continuous variation in modeling, existing VAR token reduction methods face three key limitations: heuristic stage partition, non-adaptive schedules, and limited acceleration scope, thereby leaving significant acceleration potential untapped. Since entropy variation intrinsically reflects the transition of predictive uncertainty, it offers a principled measure to capture continuous modeling variation. Therefore, we propose NOVA, a training-free token reduction acceleration framework for VAR models via entropy analysis. NOVA adaptively determines the acceleration activation scale during inference by online identifying the inflection point of scale entropy growth. Through scale-linkage and layer-linkage ratio adjustment, NOVA dynamically computes distinct token reduction ratios for each scale and layer, pruning low-entropy tokens while reusing the cache derived from the residuals at the prior scale to accelerate inference and maintain generation quality. Extensive experiments and analyses validate NOVA as a simple yet effective training-free acceleration framework. Code is available.

Modern AI image generators can create detailed, high-resolution pictures, but doing so often requires a large amount of computing time and memory. This paper introduces NOVA, a method that makes a family of image-generation models faster without requiring the models to be retrained or redesigned. The key idea is to watch how confident the model is while it is creating an image. When the model is still uncertain, NOVA keeps more computation so important visual structure is preserved. When the model becomes more confident, NOVA safely reduces repeated or less useful work and reuses helpful information from earlier steps. Unlike fixed speed-up rules, NOVA adapts to each image and to different stages of the generation process. This helps it save computation while maintaining image quality, details, and alignment with the text prompt. Experiments on several image-generation models show that NOVA can substantially reduce generation time, achieving up to 2.89× faster inference with almost no loss in quality.