Inferix: A Block-Diffusion based Next-Generation Inference Engine for World Simulation

TL;DR

Inferix introduces a block-diffusion based inference engine for world simulation, uniting diffusion and autoregressive strengths through semi-autoregressive decoding and LLM-style KV caching to support arbitrary-length, high-quality video generation. Its framework combines adaptive parallelism, robust KV management, and modular pipelines to accelerate long-form synthesis, while offering real-time streaming and built-in profiling. The LV-Bench benchmark provides a fine-grained, minute-long video evaluation suite with a dedicated dataset and a multi-metric evaluation protocol (VDE and V-Bench metrics) to quantify long-horizon stability and quality. The paper also outlines a roadmap for scalable deployment, advanced attention, and distributed inference, positioning Inferix as a community-oriented tool for advancing world-model research and long video generation. Overall, Inferix aims to lower barriers to immersive world synthesis by providing an integrated, efficient, and extensible inference ecosystem for block-diffusion models.

Abstract

World models serve as core simulators for fields such as agentic AI, embodied AI, and gaming, capable of generating long, physically realistic, and interactive high-quality videos. Moreover, scaling these models could unlock emergent capabilities in visual perception, understanding, and reasoning, paving the way for a new paradigm that moves beyond current LLM-centric vision foundation models. A key breakthrough empowering them is the semi-autoregressive (block-diffusion) decoding paradigm, which merges the strengths of diffusion and autoregressive methods by generating video tokens in block-applying diffusion within each block while conditioning on previous ones, resulting in more coherent and stable video sequences. Crucially, it overcomes limitations of standard video diffusion by reintroducing LLM-style KV Cache management, enabling efficient, variable-length, and high-quality generation. Therefore, Inferix is specifically designed as a next-generation inference engine to enable immersive world synthesis through optimized semi-autoregressive decoding processes. This dedicated focus on world simulation distinctly sets it apart from systems engineered for high-concurrency scenarios (like vLLM or SGLang) and from classic video diffusion models (such as xDiTs). Inferix further enhances its offering with interactive video streaming and profiling, enabling real-time interaction and realistic simulation to accurately model world dynamics. Additionally, it supports efficient benchmarking through seamless integration of LV-Bench, a new fine-grained evaluation benchmark tailored for minute-long video generation scenarios. We hope the community will work together to advance Inferix and foster world model exploration.

Figures (2)

• Figure 1: Architecture comparison. AR vs. Diffusion vs. Block Diffusion (Semi-AR). Block Diffusion combines the strengths of both AR and Diffusion, enabling arbitrary-length generation, KV caching, and high parallelizability within each block.
• Figure 2: Framework of Inferix. To enhance the efficiency of block diffusion models, INFERIX provides a set of interconnected components: efficient parallel strategies, block-wise KV Cache management, DAX dax2025 quantization, real-time video streaming, and fine-grained video evaluation.