Dynamic frequency modulation for controllable text-driven image generation

doi:10.1016/j.patcog.2026.114880

Dynamic frequency modulation for controllable text-driven image generation

2026-09-08

0

PRE

AI

DOI：10.1016/j.patcog.2026.114880

Original

Original request for help

Save

Abstract

En 中文

• Lower-frequency components establish the structure framework in early generation. • Higher-frequency components dominate later stages to synthesize fine-grained textures. • We propose a training-free Frequency Modulation Method with dynamic decay. • Directly manipulating noisy latent variables avoids empirical feature map selection. • Our method outperforms state-of-the-art methods in diverse generation scenarios.

Keywords:

Diffusion model

Image generation

Frequency spectrum

Structure framework

Dynamic frequency modulation for controllable text-driven image generation

Abstract

Journal

Organization

Cited Papers