NextFlow: Unified Sequential Modeling Activates Multimodal Understanding and Generation

📌 核心贡献

NextFlow是一个统一的解码器自回归Transformer模型，通过在6万亿个交错的文本-图像离散tokens上训练，实现了多模态理解和生成。其核心创新在于将统一的视觉表示融入统一的自回归架构中，并针对图像的层级特性，创造性地采用了“下一尺度预测”而非传统的“下一token预测”进行视觉生成。这使得模型能够原生支持图像编辑、交错内容和视频生成等多种高级功能。

📖 摘要

We present NextFlow, a unified decoder-only autoregressive transformer trained on 6 trillion interleaved text-image discrete tokens. By leveraging a unified vision representation within a unified autoregressive architecture, NextFlow natively activates multimodal understanding and generation capabilities, unlocking abilities of image editing, interleaved content and video generation. Motivated by the distinct nature of modalities - where text is strictly sequential and images are inherently hierarchical - we retain next-token prediction for text but adopt next-scale prediction for visual generation. This departs from traditional raster-scan methods, enabling the generation of 1024x1024 images in just 5 seconds - orders of magnitude faster than comparable AR models. We address the instabilities of multi-scale generation through a robust training recipe. Furthermore, we introduce a prefix-tuning strategy for reinforcement learning. Experiments demonstrate that NextFlow achieves state-of-the-art performance among unified models and rivals specialized diffusion baselines in visual quality.

🔍 关键信息

字段	内容
机构
发表	2026-01-05
分类	cs.CV, cs.AI
链接	arXiv

📝 我的笔记

🔗 相关论文

同方向： Transfusion, UniFluid, Show-o2