Brief

Researchers have developed PREX, a novel framework for faithful 4D video editing that addresses the challenge of preserving original regions while synthesizing new content. The method identifies and corrects an "Evidence-Role Mismatch" in existing diffusion models, which can lead to ghosting and unstable extrapolation. PREX decomposes video volumes into distinct roles (Preserve, Reveal, Expand) and uses a region-aware adapter with calibrated confidence cues, trained without paired edited videos. A new benchmark, PREBench, was also introduced to evaluate these capabilities. AI

Researchers have developed several new techniques to improve video diffusion models, focusing on efficiency and quality. One approach, LocalDPO, optimizes alignment at a localized spatio-temporal region level for better video fidelity and coherence. Another method, ARL2, replaces quadratic self-attention with a fixed-size recurrent state to achieve linear time scaling and constant memory usage, speeding up generation and reducing memory requirements. Additionally, ORBIS is an SW-HW co-designed accelerator that uses output activation for more accurate inter-token similarity, leading to higher token reduction ratios and significant speedup and energy reduction. Finally, Bernini unifies multimodal large language models (MLLMs) with diffusion models, using MLLMs for semantic planning and diffusion models for pixel rendering, achieving state-of-the-art performance in video generation and editing. AI