DC-AR: Efficient Masked Autoregressive Image Generation with Deep Compression Hybrid Tokenizer

2025-07-07

AI Summary

Existing masked autoregressive (AR) image generation models lag behind diffusion models in both quality and efficiency due to limitations of conventional tokenizers. To address this, we propose an efficient text-to-image generation framework. Our core innovation is the Deeply Compressed Hybrid Tokenizer (DC-HT), enabling 32× spatial compression, high-fidelity reconstruction, and cross-resolution generalization. We further introduce a two-stage AR generation paradigm: first modeling the image skeleton using discrete structural tokens, then refining fine details via residual tokens. Built upon MaskGIT, our method achieves 5.49 gFID and 0.69 GenEval on MJHQ-30K—setting new state-of-the-art performance. It also delivers 1.5–7.9× higher throughput and 2.0–3.5× lower latency compared to prior AR approaches. Collectively, these advances establish a new Pareto-optimal trade-off between generation quality, speed, and computational efficiency among autoregressive image synthesis methods.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Generation

Application Category

• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications

Abstract

We introduce DC-AR, a novel masked autoregressive (AR) text-to-image generation framework that delivers superior image generation quality with exceptional computational efficiency. Due to the tokenizers' limitations, prior masked AR models have lagged behind diffusion models in terms of quality or efficiency. We overcome this limitation by introducing DC-HT - a deep compression hybrid tokenizer for AR models that achieves a 32x spatial compression ratio while maintaining high reconstruction fidelity and cross-resolution generalization ability. Building upon DC-HT, we extend MaskGIT and create a new hybrid masked autoregressive image generation framework that first produces the structural elements through discrete tokens and then applies refinements via residual tokens. DC-AR achieves state-of-the-art results with a gFID of 5.49 on MJHQ-30K and an overall score of 0.69 on GenEval, while offering 1.5-7.9x higher throughput and 2.0-3.5x lower latency compared to prior leading diffusion and autoregressive models.

Problem

• Overcome tokenizer limitations in autoregressive image generation
• Improve image generation quality and computational efficiency
• Achieve high compression ratio with reconstruction fidelity

Innovation

• Deep Compression Hybrid Tokenizer (DC-HT)
• Hybrid masked autoregressive framework
• 32x spatial compression ratio