Image Diffusion Preview with Consistency Solver
Abstract
Diffusion Preview uses ConsistencySolver, a high-order trainable solver, to improve quality and consistency in low-step image generation, enhancing interactive user experiences.
The slow inference process of image diffusion models significantly degrades interactive user experiences. To address this, we introduce Diffusion Preview, a novel paradigm employing rapid, low-step sampling to generate preliminary outputs for user evaluation, deferring full-step refinement until the preview is deemed satisfactory. Existing acceleration methods, including training-free solvers and post-training distillation, struggle to deliver high-quality previews or ensure consistency between previews and final outputs. We propose ConsistencySolver derived from general linear multistep methods, a lightweight, trainable high-order solver optimized via Reinforcement Learning, that enhances preview quality and consistency. Experimental results demonstrate that ConsistencySolver significantly improves generation quality and consistency in low-step scenarios, making it ideal for efficient preview-and-refine workflows. Notably, it achieves FID scores on-par with Multistep DPM-Solver using 47% fewer steps, while outperforming distillation baselines. Furthermore, user studies indicate our approach reduces overall user interaction time by nearly 50% while maintaining generation quality. Code is available at https://github.com/G-U-N/consolver.
Community
The slow inference process of image diffusion models significantly degrades interactive user experiences. We introduce Diffusion Preview, a novel preview-and-refine paradigm that generates rapid, low-step preliminary outputs for user evaluation, deferring full-step refinement until the preview is deemed satisfactory. This workflow enables users to quickly iterate through different prompts or random seeds with minimal computational cost, only triggering expensive full-step sampling when a preview meets their expectations.
Diffusion Preview framework: Fast preview generation followed by full-step refinement.
To achieve high-quality and consistent previews, we propose ConsistencySolver, a learnable high-order ODE solver derived from Linear Multistep Methods and optimized via Reinforcement Learning. Unlike existing training-free solvers that rely on rigid numerical schemes or distillation methods that sacrifice consistency, ConsistencySolver dynamically adapts its integration strategy to maximize alignment between low-step previews and high-step reference generations, ensuring previews serve as reliable proxies for final outputs.
Overview of our RL framework for optimizing a learnable ODE solver in diffusion sampling.
Empirical validation demonstrates that ConsistencySolver significantly outperforms training-free ODE solvers (e.g., DDIM, UniPC, Multistep DPM), distillation-based methods (e.g., LCM, PCM, DMD2), and distillation-based solvers (e.g., AMED) across both consistency metrics and FID scores.
Quantitative Results on Stable Diffusion v1-5 for Text-to-Image Generation.
This is an automated message from the Librarian Bot. I found the following papers similar to this paper.
The following papers were recommended by the Semantic Scholar API
- BADiff: Bandwidth Adaptive Diffusion Model (2025)
- OmniRefiner: Reinforcement-Guided Local Diffusion Refinement (2025)
- ProxT2I: Efficient Reward-Guided Text-to-Image Generation via Proximal Diffusion (2025)
- AutoRefiner: Improving Autoregressive Video Diffusion Models via Reflective Refinement Over the Stochastic Sampling Path (2025)
- Flash-DMD: Towards High-Fidelity Few-Step Image Generation with Efficient Distillation and Joint Reinforcement Learning (2025)
- PaCo-RL: Advancing Reinforcement Learning for Consistent Image Generation with Pairwise Reward Modeling (2025)
- Masked Auto-Regressive Variational Acceleration: Fast Inference Makes Practical Reinforcement Learning (2025)
Please give a thumbs up to this comment if you found it helpful!
If you want recommendations for any Paper on Hugging Face checkout this Space
You can directly ask Librarian Bot for paper recommendations by tagging it in a comment:
@librarian-bot
recommend
Models citing this paper 0
No model linking this paper
Datasets citing this paper 0
No dataset linking this paper
Spaces citing this paper 0
No Space linking this paper
Collections including this paper 0
No Collection including this paper