Search papers, labs, and topics across Lattice.
The paper analyzes the spectral properties of cross-attention in diffusion models, revealing a consistent coarse-to-fine frequency progression during denoising. Based on this analysis, they introduce Attention Frequency Modulation (AFM), a training-free inference-time method that manipulates cross-attention logits in the Fourier domain by reweighting frequency bands. AFM allows for controllable visual edits in Stable Diffusion by biasing the spatial scale of token competition, while preserving semantic alignment.
Steer Stable Diffusion's attention like an equalizer, sculpting image details without retraining by simply tweaking the frequency spectrum of cross-attention.
Cross-attention is the primary interface through which text conditions latent diffusion models, yet its step-wise multi-resolution dynamics remain under-characterized, limiting principled training-free control. We cast diffusion cross-attention as a spatiotemporal signal on the latent grid by summarizing token-softmax weights into token-agnostic concentration maps and tracking their radially binned Fourier power over denoising. Across prompts and seeds, encoder cross-attention exhibits a consistent coarse-to-fine spectral progression, yielding a stable time-frequency fingerprint of token competition. Building on this structure, we introduce Attention Frequency Modulation (AFM), a plug-and-play inference-time intervention that edits token-wise pre-softmax cross-attention logits in the Fourier domain: low- and high-frequency bands are reweighted with a progress-aligned schedule and can be adaptively gated by token-allocation entropy, before the token softmax. AFM provides a continuous handle to bias the spatial scale of token-competition patterns without retraining, prompt editing, or parameter updates. Experiments on Stable Diffusion show that AFM reliably redistributes attention spectra and produces substantial visual edits while largely preserving semantic alignment. Finally, we find that entropy mainly acts as an adaptive gain on the same frequency-based edit rather than an independent control axis.
