Efficient One-Step Diffusion Restoration Model with Compact Token Compression and Linear Attention

2026-05-22

AI Summary

Existing real-world image super-resolution methods suffer from high computational overhead and deployment challenges due to dense latent representations and quadratic-complexity attention mechanisms. This work proposes SANA-SR, which for the first time integrates a high-ratio (32×) compressed autoencoder with a linear-complexity DiT attention mechanism to construct a compact token representation. The approach further incorporates LoRA fine-tuning and model pruning to substantially reduce computational redundancy. SANA-SR achieves state-of-the-art or competitive performance across multiple benchmarks while producing sharper and more photorealistic textures. After pruning, the model requires only 0.019 seconds of inference time, 407.95G MACs, and 344M parameters, demonstrating strong potential for mobile deployment.

Abstract

Real-world image super-resolution aims to recover high-quality images from complex and unknown real-world degradations. However, existing generative Real-ISR methods largely inherit the dense latent representations and quadratic-cost global modeling paradigm developed for high-resolution image synthesis, causing computation, memory usage, and inference latency to scale unfavorably with resolution and thus limiting practical deployment. We argue that the key bottleneck lies not in insufficient restoration priors, but in excessive token redundancy and costly token interactions during high-resolution restoration. Motivated by this observation, we revisit Real-ISR from the perspectives of compact latent representation and linear-complexity modeling, and propose SANA-SR, an efficient one-step restoration framework. Specifically, SANA-SR employs a deep compression autoencoder with a 32x compression ratio to drastically reduce latent tokens while preserving restoration-relevant structures and textures. On top of this compact latent space, we introduce a linear-attention DiT with LoRA fine-tuning, enabling efficient high-resolution restoration with linear-complexity token mixing. Extensive experiments on all benchmark datasets demonstrate that SANA-SR achieves highly competitive and often superior quantitative performance against existing methods, while restoring clearer and more realistic textures. Moreover, after pruning, the deployed model runs in 0.019s with 407.95G MACs and 344M parameters, highlighting its strong potential for practical mobile deployment.

Problem

• Real-world image super-resolution
• computational efficiency
• token redundancy
• high-resolution restoration
• model deployment

Innovation

• compact token compression
• linear attention
• one-step diffusion
• real-world super-resolution
• efficient restoration