TFG: Unified Training-Free Guidance for Diffusion Models

TL;DR

This paper introduces TFG, a unified, training-free guidance framework for diffusion models that improves sample quality across various tasks through a theoretical and empirical approach, enabling easier application to new problems.

Contribution

It presents a novel, unified framework for training-free guidance in diffusion models, including a hyper-parameter search strategy and extensive benchmarking across multiple models and tasks.

Findings

Achieved an average performance improvement of 8.5% across benchmarks.

Unified existing methods into a single algorithm-agnostic framework.

Provided a comprehensive benchmark suite for training-free diffusion guidance.

Abstract

Given an unconditional diffusion model and a predictor for a target property of interest (e.g., a classifier), the goal of training-free guidance is to generate samples with desirable target properties without additional training. Existing methods, though effective in various individual applications, often lack theoretical grounding and rigorous testing on extensive benchmarks. As a result, they could even fail on simple tasks, and applying them to a new problem becomes unavoidably difficult. This paper introduces a novel algorithmic framework encompassing existing methods as special cases, unifying the study of training-free guidance into the analysis of an algorithm-agnostic design space. Via theoretical and empirical investigation, we propose an efficient and effective hyper-parameter searching strategy that can be readily applied to any downstream task. We systematically benchmark across 7 diffusion models on 16 tasks with 40 targets, and improve performance by 8.5% on average. Our framework and benchmark offer a solid foundation for conditional generation in a training-free manner.