A-FloPS: Accelerating Diffusion Models via Adaptive Flow Path Sampler

Diffusion models deliver state-of-the-art generative performance across diverse modalities but remain computationally expensive due to their inherently iterative sampling process. Existing training-free acceleration methods typically improve numerical solvers for the reverse-time ODE, yet their effectiveness is fundamentally constrained by the inefficiency of the underlying sampling trajectories. We propose A-FloPS (Adaptive Flow Path Sampler), a principled, training-free framework that reparameterizes the sampling trajectory of any pre-trained diffusion model into a flow-matching form and augments it with an adaptive velocity decomposition. The reparameterization analytically maps diffusion scores to flow-compatible velocities, yielding integration-friendly trajectories without retraining. The adaptive mechanism further factorizes the velocity field into a linear drift term and a residual component whose temporal variation is actively suppressed, restoring the accuracy benefits of high-order integration even in extremely low-NFE regimes. Extensive experiments on conditional image generation and text-to-image synthesis show that A-FloPS consistently outperforms state-of-the-art training-free samplers in both sample quality and efficiency. Notably, with as few as $5$ function evaluations, A-FloPS achieves substantially lower FID and generates sharper, more coherent images. The adaptive mechanism also improves native flow-based generative models, underscoring its generality. These results position A-FloPS as a versatile and effective solution for high-quality, low-latency generative modeling.

💡 Research Summary

The paper introduces A‑FloPS (Adaptive Flow Path Sampler), a training‑free acceleration framework for diffusion models that re‑parameterizes any pre‑trained diffusion sampler into a flow‑matching (FM) trajectory and then adaptively decomposes the resulting velocity field to restore the benefits of high‑order numerical integration even under extremely low numbers of function evaluations (NFE).

Motivation and Background
Diffusion models achieve state‑of‑the‑art generation quality but require hundreds of neural network evaluations during sampling, which makes real‑time deployment impractical. Existing training‑free accelerators (DDIM, DPM‑Solver, UniPC) focus on improving the ODE solver but are limited by the inherent inefficiency of the original sampling path. Flow‑matching models, by contrast, learn deterministic velocity fields that transport a simple base distribution to the data distribution, yielding smoother, more regular trajectories that are easier to integrate. However, FM models need dedicated training and are not directly applicable to arbitrary diffusion models with different noise schedules.

Key Contributions

1. Diffusion‑to‑Flow Re‑parameterization (FloPS) – The authors derive an analytical bijection (Theorem 1) that maps a diffusion model’s score function (s_\theta(x,\tau)) to an FM velocity field (v^*(x,t)). The mapping is given by
   \