MosaicDiff: Training-free Structural Pruning for Diffusion Model Acceleration Reflecting Pretraining Dynamics

• URL: http://arxiv.org/abs/2510.11962v1
• Date: Mon, 13 Oct 2025 21:51:04 GMT
• Title: MosaicDiff: Training-free Structural Pruning for Diffusion Model Acceleration Reflecting Pretraining Dynamics
• Authors: Bowei Guo, Shengkun Tang, Cong Zeng, Zhiqiang Shen,
• Abstract summary: We introduce a novel framework called MosaicDiff that aligns diffusion pretraining dynamics with post-training sampling acceleration.<n>Our method achieves significant speed-ups in sampling without compromising output quality.
• Score: 34.69318408652807
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Diffusion models are renowned for their generative capabilities, yet their pretraining processes exhibit distinct phases of learning speed that have been entirely overlooked in prior post-training acceleration efforts in the community. In this study, we introduce a novel framework called MosaicDiff that aligns diffusion pretraining dynamics with post-training sampling acceleration via trajectory-aware structural pruning. Our approach leverages the observation that the middle, fast-learning stage of diffusion pretraining requires more conservative pruning to preserve critical model features, while the early and later, slow-learning stages benefit from a more aggressive pruning strategy. This adaptive pruning mechanism is the first to explicitly mirror the inherent learning speed variations of diffusion pretraining, thereby harmonizing the model's inner training dynamics with its accelerated sampling process. Extensive experiments on DiT and SDXL demonstrate that our method achieves significant speed-ups in sampling without compromising output quality, outperforming previous state-of-the-art methods by large margins, also providing a new viewpoint for more efficient and robust training-free diffusion acceleration.

Related papers

• Diff-ES: Stage-wise Structural Diffusion Pruning via Evolutionary Search [40.67449277026597]
  We introduce textbfDiff-ES, a stage-wise structural textbfDiff-usion pruning framework via textbfEvolutionary textbfSearch.<n>Our framework naturally integrates with existing structured pruning methods for diffusion models including depth and width pruning.<n>Experiments on DiT and SDXL demonstrate that Diff-ES consistently achieves wall-clock speedups while incurring minimal degradation in generation quality.
  arXiv  Detail & Related papers  (2026-03-05T12:18:40Z)
• Analyzing and Improving Fast Sampling of Text-to-Image Diffusion Models [32.70019265781621]
  Text-to-image diffusion models have achieved unprecedented success but still struggle to produce high-quality images under limited sampling budgets.<n>We propose constant total rotation schedule (TORS) as a scheduling strategy that ensures uniform geometric variation along the sampling trajectory.<n>TORS outperforms previous training-free acceleration methods and produces high-quality images with 10 sampling steps on Flux.1-Dev and Stable Diffusion 3.5.
  arXiv  Detail & Related papers  (2026-02-28T18:09:44Z)
• Boosting Fidelity for Pre-Trained-Diffusion-Based Low-Light Image Enhancement via Condition Refinement [63.54516423266521]
  Pre-Trained Diffusion-Based (PTDB) methods often sacrifice content fidelity to attain higher perceptual realism.<n>We propose a novel optimization strategy for conditioning in pre-trained diffusion models, enhancing fidelity while preserving realism and aesthetics.<n>Our approach is plug-and-play, seamlessly integrating into existing diffusion networks to provide more effective control.
  arXiv  Detail & Related papers  (2025-10-20T02:40:06Z)
• Pretrained Diffusion Models Are Inherently Skipped-Step Samplers [4.858858247064974]
  We provide a confirmative answer and introduce skipped-step sampling, a mechanism that bypasses multiple intermediate denoising steps in the iterative generation process.<n>We demonstrate that this skipped-step sampling mechanism is derived from the same training objective as the standard diffusion model.<n>We propose an enhanced generation method by integrating our accelerated sampling technique with DDIM.
  arXiv  Detail & Related papers  (2025-08-21T04:45:13Z)
• FlowMo: Variance-Based Flow Guidance for Coherent Motion in Video Generation [51.110607281391154]
  FlowMo is a training-free guidance method for enhancing motion coherence in text-to-video models.<n>It estimates motion coherence by measuring the patch-wise variance across the temporal dimension and guides the model to reduce this variance dynamically during sampling.
  arXiv  Detail & Related papers  (2025-06-01T19:55:33Z)
• Mean Flows for One-step Generative Modeling [64.4997821467102]
  We propose a principled and effective framework for one-step generative modeling.<n>A well-defined identity between average and instantaneous velocities is derived and used to guide neural network training.<n>Our method, termed the MeanFlow model, is self-contained and requires no pre-training, distillation, or curriculum learning.
  arXiv  Detail & Related papers  (2025-05-19T17:59:42Z)
• Dynamical Diffusion: Learning Temporal Dynamics with Diffusion Models [71.63194926457119]
  We introduce Dynamical Diffusion (DyDiff), a theoretically sound framework that incorporates temporally aware forward and reverse processes.<n>Experiments across scientifictemporal forecasting, video prediction, and time series forecasting demonstrate that Dynamical Diffusion consistently improves performance in temporal predictive tasks.
  arXiv  Detail & Related papers  (2025-03-02T16:10:32Z)
• Adaptive Non-Uniform Timestep Sampling for Diffusion Model Training [4.760537994346813]
  As data distributions grow more complex, training diffusion models to convergence becomes increasingly intensive. We introduce a non-uniform timestep sampling method that prioritizes these more critical timesteps. Our method shows robust performance across various datasets, scheduling strategies, and diffusion architectures.
  arXiv  Detail & Related papers  (2024-11-15T07:12:18Z)
• A Closer Look at Time Steps is Worthy of Triple Speed-Up for Diffusion Model Training [53.93563224892207]
  We introduce a novel speed-up method for diffusion model training, called, which is based on a closer look at time steps.<n>As a plug-and-play and architecture-agnostic approach, SpeeD consistently achieves 3-times acceleration across various diffusion architectures, datasets, and tasks.
  arXiv  Detail & Related papers  (2024-05-27T17:51:36Z)
• Preparing Lessons for Progressive Training on Language Models [75.88952808979087]
  The rapid progress of Transformers in artificial intelligence has come at the cost of increased resource consumption and greenhouse gas emissions. We propose Apollo, which preptextbfares lessons for extextbfpanding textbfoperations by textbflayer functitextbfonality during training of low layers. Experiments demonstrate that Apollo achieves state-of-the-art acceleration ratios, even rivaling methods using pretrained models.
  arXiv  Detail & Related papers  (2024-01-17T13:04:14Z)
• Fast Diffusion Model [122.36693015093041]
  Diffusion models (DMs) have been adopted across diverse fields with their abilities in capturing intricate data distributions. In this paper, we propose a Fast Diffusion Model (FDM) to significantly speed up DMs from a DM optimization perspective.
  arXiv  Detail & Related papers  (2023-06-12T09:38:04Z)
• DYffusion: A Dynamics-informed Diffusion Model for Spatiotemporal Forecasting [18.86526240105348]
  We propose an approach for efficiently training diffusion models for probabilistic forecasting. We train a time-conditioned interpolator and a forecaster network that mimic the forward and reverse processes of standard diffusion models. Our approach performs competitively on probabilistic forecasting of complex dynamics in sea surface temperatures, Navier-Stokes flows, and flows spring systems.
  arXiv  Detail & Related papers  (2023-06-03T02:46:31Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.