Unlocking Few-Step Diffusion for Faithful Previews

2026-09-28

AI Summary

This study addresses the output distortion and cumulative sampling latency inherent in few-step generators within diffusion model workflows. We propose a training-free input correction method that generalizes across computational budgets. By leveraging the latent reconstruction capability of a frozen few-step sampler, our approach jointly optimizes the initial noise and denoising updates via endpoint-supervised learning, enabling high-fidelity rapid previews for efficient candidate image screening. Experimental results demonstrate that the proposed method reduces reconstruction mean squared error by 53%–78% compared to the LD3 baseline, significantly improving reference fidelity and candidate ranking accuracy. These findings establish a new paradigm for the efficient deployment of diffusion models.

Abstract

Sampling latency compounds in diffusion workflows, where users generate and discard many candidates before keeping one. Surprisingly, the poor outputs of standard few-step samplers do not reflect a lack of reconstruction capacity: by optimizing only the initial noise, frozen 3-4-step samplers can closely reproduce their corresponding full-step outputs. Building on this finding, we learn corrections to the initial noise and denoising updates using endpoint supervision, improving correspondence with full-step outputs generated from the same noise and prompt. The resulting previews allow users to screen candidates cheaply and reserve full-step generation for promising ones. Input correction also transfers across sampling budgets without retraining. Experiments show substantial improvements in reference fidelity, including 53-78% lower reconstruction MSE than retrained LD3 on unconditional benchmarks, alongside improved ranking preservation and candidate selection on SD1.5, SDXL, and FLUX.1-dev.

Problem

• diffusion models
• few-step sampling
• sampling latency
• faithful previews
• candidate screening

Innovation

• Few-Step Diffusion
• Endpoint Supervision
• Initial Noise Optimization
• Preview Generation
• Sampling Budget Transfer