Drag Your Noise: Interactive Point-based Editing via Diffusion Semantic Propagation

TL;DR

DragNoise introduces a robust, efficient point-based editing method for diffusion models that leverages U-Net's semantic features, enabling single-step edits with improved content preservation and over 50% faster performance.

Contribution

We propose DragNoise, a novel diffusion editing approach that uses predicted noise as a semantic editor, avoiding latent map retracing and enhancing editing stability and speed.

Findings

DragNoise outperforms DragDiffusion in control and semantic retention.

It reduces optimization time by over 50%.

It enables single-step, stable diffusion editing.

Abstract

Point-based interactive editing serves as an essential tool to complement the controllability of existing generative models. A concurrent work, DragDiffusion, updates the diffusion latent map in response to user inputs, causing global latent map alterations. This results in imprecise preservation of the original content and unsuccessful editing due to gradient vanishing. In contrast, we present DragNoise, offering robust and accelerated editing without retracing the latent map. The core rationale of DragNoise lies in utilizing the predicted noise output of each U-Net as a semantic editor. This approach is grounded in two critical observations: firstly, the bottleneck features of U-Net inherently possess semantically rich features ideal for interactive editing; secondly, high-level semantics, established early in the denoising process, show minimal variation in subsequent stages. Leveraging these insights, DragNoise edits diffusion semantics in a single denoising step and efficiently propagates these changes, ensuring stability and efficiency in diffusion editing. Comparative experiments reveal that DragNoise achieves superior control and semantic retention, reducing the optimization time by over 50% compared to DragDiffusion. Our codes are available at https://github.com/haofengl/DragNoise.