Multi-Resolution Flow Matching: Training-Free Diffusion Acceleration via Staged Sampling

2026-07-01

AI Summary

Existing training-free multi-resolution diffusion acceleration methods operating in latent space often introduce blurriness or artifacts. This work proposes MrFlow, a training-free multi-resolution acceleration strategy that progressively infers from low to high resolution to accelerate pretrained flow-matching models. Its key innovation lies in the first-time integration of lightweight GAN-based super-resolution with low-intensity noise injection directly in pixel space, enabling training- and runtime-recognition-free high-frequency resampling. MrFlow is orthogonal to and can be combined with other acceleration techniques. On FLUX.1-dev and Qwen-Image, MrFlow achieves a 10× end-to-end speedup with less than 1% degradation in OneIG score; when further integrated with timestep distillation, it attains up to 25× acceleration.

Abstract

Hardware-agnostic strategies for accelerating text-to-image diffusion, such as timestep distillation and feature caching, can reduce inference time without custom kernels or system-level optimization. Among them, multi-resolution generation strategies have recently received broad attention, attaining more than 5x speedup without any training. However, the design of performing upsampling in the latent space, together with the selective modification of partial regions, causes these methods to exhibit noticeable blurring or artifacts. To this end, we propose MrFlow, a training-free multi-resolution acceleration strategy for pretrained flow-matching models built upon a staged low-to-high-resolution pipeline. MrFlow first rapidly generates the main structure at low resolution, then performs super-resolution in the pixel space using a lightweight pretrained GAN-based model, subsequently injects low-strength noise to enable high-frequency resampling, and finally refines the details at high resolution. Quantitative and qualitative results on FLUX.1-dev and Qwen-Image show that MrFlow exploits the quadratic token reduction and reduced step requirement of low-resolution sampling to achieve 10x end-to-end acceleration while keeping OneIG within a 1% gap relative to that before acceleration, significantly surpassing other training-free acceleration strategies, and requiring no training or runtime dynamic identification whatsoever. MrFlow can further be directly combined orthogonally with pre-trained timestep distillation strategies, achieving even higher generation acceleration of up to 25x.

Problem

• multi-resolution generation
• diffusion acceleration
• artifacts
• blurring
• training-free

Innovation

• flow matching
• multi-resolution generation
• training-free acceleration
• pixel-space super-resolution
• diffusion acceleration