Unified Multimodal Autoregressive Modeling with Shared Context-Visual Tokenizer is Key to Unification
2026-06-16 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors present UniAR, a system that unifies how computers understand and create images by using a single kind of visual token. Unlike previous methods that use two different ways to represent images, UniAR uses one visual tokenizer, making it easier for the model to work with images consistently. They improve image quality and speed by combining features from different layers and predicting multiple parts of the image code simultaneously. After training on large datasets, UniAR performs very well at both generating and editing images, while still being good at understanding tasks involving images and text together.
unified multimodal modelingvisual tokenizerautoregressive modelfeature fusionbitwise quantizationvisual vocabularydiffusion-based decoderimage generationreinforcement learningmultimodal understanding
Authors
Wujian Peng, Lingchen Meng, Yuxuan Cai, Xianwei Zhuang, Yuhuan Yang, Rongyao Fang, Chenfei Wu, Junyang Lin, Zuxuan Wu, Shuai Bai
Abstract
Unified Multimodal Modeling aims to integrate visual understanding and generation within a single system. However, existing approaches typically rely on two disparate visual tokenizers, which splits the representation space and hinders truly unified modeling. We propose UniAR, a unified autoregressive framework where a single discrete visual tokenizer serves as the key bridge between understanding and generation, enabling a shared context in which the model can directly interpret its own generated visual tokens without additional re-encoding. UniAR adapts a pretrained vision encoder with multi-level feature fusion and a lookup-free bitwise quantization scheme, preserving both high-level semantics and low-level details while scaling the effective visual vocabulary at minimal cost. Building on this, the unified autoregressive model adopts parallel-bitwise-prediction to jointly predict spatially grouped, multi-level visual codes, substantially reducing visual sequence length and accelerating generation. Finally, a diffusion-based visual decoder operates on discrete visual tokens to decode high-fidelity images. Through large-scale pre-training, followed by supervised fine-tuning and reinforcement learning, UniAR achieves state-of-the-art performance on image generation and image editing while remaining competitive on multimodal understanding benchmarks. The project page is available at https://sharelab-sii.github.io/uniar-web.