Image-Free Timestep Distillation via Continuous-Time Consistency with Trajectory-Sampled Pairs

TL;DR

This paper introduces TBCM, a novel self-contained distillation method for diffusion models that extracts latent representations from the teacher's generation trajectory, significantly improving efficiency and scalability without relying on external data.

Contribution

The paper proposes TBCM, a trajectory-based distillation approach that eliminates the need for external datasets and reduces computational resources, enhancing diffusion model training.

Findings

Achieves 6.52 FID and 28.08 CLIP scores on MJHQ-30k with one-step generation.

Reduces training time by approximately 40% compared to Sana-Sprint.

Saves substantial GPU memory while maintaining high generation quality.

Abstract

Timestep distillation is an effective approach for improving the generation efficiency of diffusion models. The Consistency Model (CM), as a trajectory-based framework, demonstrates significant potential due to its strong theoretical foundation and high-quality few-step generation. Nevertheless, current continuous-time consistency distillation methods still rely heavily on training data and computational resources, hindering their deployment in resource-constrained scenarios and limiting their scalability to diverse domains. To address this issue, we propose Trajectory-Backward Consistency Model (TBCM), which eliminates the dependence on external training data by extracting latent representations directly from the teacher model's generation trajectory. Unlike conventional methods that require VAE encoding and large-scale datasets, our self-contained distillation paradigm significantly improves both efficiency and simplicity. Moreover, the trajectory-extracted samples naturally bridge the distribution gap between training and inference, thereby enabling more effective knowledge transfer. Empirically, TBCM achieves 6.52 FID and 28.08 CLIP scores on MJHQ-30k under one-step generation, while reducing training time by approximately 40% compared to Sana-Sprint and saving a substantial amount of GPU memory, demonstrating superior efficiency without sacrificing quality. We further reveal the diffusion-generation space discrepancy in continuous-time consistency distillation and analyze how sampling strategies affect distillation performance, offering insights for future distillation research. GitHub Link: https://github.com/hustvl/TBCM.