GLASS Flows: Transition Sampling for Alignment of Flow and Diffusion Models

2025-09-29

AI Summary

Existing reward alignment algorithms rely on SDE-based sampling to realize Markov transitions, improving alignment between flow matching and diffusion models but suffering from significantly lower computational efficiency than ODE-based sampling—introducing an inherent trade-off between stochasticity and inference speed. This work proposes GLASS Flows, a nested flow-matching framework that implicitly constructs “flow matching within flow matching” inside pre-trained models, enabling extraction of inner structures without retraining. It thus unifies the computational efficiency of ODE solvers with the stochastic evolution capability of SDEs. Integrated with Feynman–Kac steering for reward-aligned optimization, GLASS Flows offers a plug-and-play, inference-time extension. Evaluated on large-scale text-to-image generation, it achieves state-of-the-art performance, substantially improving both sample quality and sampling efficiency.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Search and Optimization: Sampling/Simulation-based Search
• Machine Learning: Deep Neural Architectures and Foundation Models

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Large language models for search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

The performance of flow matching and diffusion models can be greatly improved at inference time using reward alignment algorithms, yet efficiency remains a major limitation. While several algorithms were proposed, we demonstrate that a common bottleneck is the sampling method these algorithms rely on: many algorithms require to sample Markov transitions via SDE sampling, which is significantly less efficient and often less performant than ODE sampling. To remove this bottleneck, we introduce GLASS Flows, a new sampling paradigm that simulates a "flow matching model within a flow matching model" to sample Markov transitions. As we show in this work, this "inner" flow matching model can be retrieved from a pre-trained model without any re-training, combining the efficiency of ODEs with the stochastic evolution of SDEs. On large-scale text-to-image models, we show that GLASS Flows eliminate the trade-off between stochastic evolution and efficiency. Combined with Feynman-Kac Steering, GLASS Flows improve state-of-the-art performance in text-to-image generation, making it a simple, drop-in solution for inference-time scaling of flow and diffusion models.

Problem

• Improving efficiency of reward alignment in flow/diffusion models
• Enabling stochastic transitions without SDE sampling limitations
• Combining ODE efficiency with SDE-like stochastic evolution

Innovation

• GLASS Flows samples Markov transitions via inner flow matching
• Retrieves inner flow matching model without retraining
• Combines ODE efficiency with SDE stochastic evolution