D-Flow: Differentiating through Flows for Controlled Generation
This work addresses controlled generation for applications in inverse problems and conditional generation, offering a tool that avoids task-specific retraining, though it is incremental as it builds on existing Diffusion/FM models.
The authors tackled the problem of controlling generation outcomes in Diffusion and Flow-Matching models without retraining, introducing D-Flow, a framework that optimizes the source noise point by differentiating through the flow, achieving state-of-the-art performance in tasks like image and audio inverse problems and conditional molecule generation.
Taming the generation outcome of state of the art Diffusion and Flow-Matching (FM) models without having to re-train a task-specific model unlocks a powerful tool for solving inverse problems, conditional generation, and controlled generation in general. In this work we introduce D-Flow, a simple framework for controlling the generation process by differentiating through the flow, optimizing for the source (noise) point. We motivate this framework by our key observation stating that for Diffusion/FM models trained with Gaussian probability paths, differentiating through the generation process projects gradient on the data manifold, implicitly injecting the prior into the optimization process. We validate our framework on linear and non-linear controlled generation problems including: image and audio inverse problems and conditional molecule generation reaching state of the art performance across all.