One-Step Distillation of Discrete Diffusion Image Generators via Fixed-Point Iteration

TL;DR

This paper introduces Fixed-Point Distillation (FPD), a novel end-to-end method for one-step discrete diffusion image generation that improves speed and quality by refining a single draft with a teacher model.

Contribution

FPD constructs local correction targets through partial corruption and refinement, using a continuous feature space and a straight-through estimator for effective training and inference.

Findings

FPD achieves competitive visual fidelity in a single inference step.

It narrows the gap between one-step and multi-step diffusion models.

Outperforms existing discrete distillation baselines.

Abstract

Discrete diffusion models excel at visual synthesis but rely on slow, iterative decoding. Existing single-step distillation methods attempt to bypass this bottleneck, either by training auxiliary score networks that effectively double compute, or by introducing specialized parameterizations and multi-stage pipelines that fragment optimization. In this paper, we introduce Fixed-Point Distillation (FPD), an end-to-end framework that constructs local correction targets by partially corrupting the student's one-step draft and refining it with a single teacher step. To compute the training objective in a semantically meaningful space, we lift discrete tokens into continuous features and apply a multi-bandwidth drift loss that iteratively accumulates these corrections. To backpropagate through the discrete bottleneck, we employ a straight-through estimator that feeds exact hard-sampled tokens to the teacher and decoder during the forward pass, ensuring that training and inference operate on the same codebook manifold, while routing continuous gradients back to the student logits. This fully differentiable pathway additionally accommodates an optional unconditional adversarial objective to enhance perceptual realism. Evaluations on both class- and text-conditional generation validate the effectiveness of our framework. FPD achieves competitive visual fidelity and structural alignment within a single inference step, narrowing the gap to multi-step teachers while outperforming existing discrete distillation baselines.