WaiT for the Signal: Simple Frequency-Aware Flow-Matching
Abstract
As image generation models scale to ever higher resolutions, global coherence, local detail, and texture fidelity become critical axes for generation quality. However, standard flow matching treats all spatial frequencies uniformly, ignoring the natural frequency hierarchy where high-frequency bands become indistinguishable from pure noise far earlier than coarse structures. We introduce WaiT, a Wavelet-aware image Transformer that decomposes generation into coarse and fine bands via lossless wavelets. True to its name, the high-frequency bands wait for the signal: staying pure noise until coarse structure has emerged, then joining the flow for joint refinement. Since standard FID discards fine-grained detail through aggressive downsampling, we introduce a more stringent three-axis evaluation protocol to assess quality at native resolution. On ImageNet 512x512, WaiT achieves a pixel-space FID of 1.43 and is Pareto-optimal across all three axes, reducing sampling compute by up to 50%. With our largest 2B model, we set a new state-of-the-art FID of 1.3 for pixel-space models on ImageNet 512 resolution. Our formulation outperforms even the strongest latent-space models on texture fidelity, and scales seamlessly to high-resolution OpenImages and to video generation, achieving a state-of-the-art FVD of 0.84 on Kinetics-600 with no algorithmic modifications.
Related
- The maximum size of simple solid matching covered graphs
- Irregular Sampling and Spectral Estimation: Statistical Analysis and Frequency-Refined Orthogonal Matching Pursuit
- Reconstruction-Aware Distribution Matching for Inverse Problems with Unpaired Data
- TAVRUM: Feasibility-First Contract Matching for Fragmentation-Aware Resource Allocation in IoT–Fog Networks
- EVTradeMatch: A Mobility-Aware Multi-Objective Matching Framework for EV--EV Energy Trading
BibTeX
@misc{pavasovic2026wait,
title = {{WaiT for the Signal: Simple Frequency-Aware Flow-Matching}},
author = {Krunoslav Lehman Pavasovic and Théophane Vallaeys and Stéphane Mallat and Giulio Biroli and Luke Zettlemoyer and Brian Karrer and Jakob Verbeek},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2607.28760},
eprint = {2607.28760},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2607.28760},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - WaiT for the Signal: Simple Frequency-Aware Flow-Matching
AU - Krunoslav Lehman Pavasovic
AU - Théophane Vallaeys
AU - Stéphane Mallat
AU - Giulio Biroli
AU - Luke Zettlemoyer
AU - Brian Karrer
AU - Jakob Verbeek
PY - 2026
DA - 2026/10/30
PB - arXiv
DO - 10.48550/arxiv.2607.28760
UR - https://arxiv.org/abs/2607.28760
AB - As image generation models scale to ever higher resolutions, global coherence, local detail, and texture fidelity become critical axes for generation quality. However, standard flow matching treats all spatial frequencies uniformly, ignoring the natural frequency hierarchy where high-frequency bands become indistinguishable from pure noise far earlier than coarse structures. We introduce WaiT, a Wavelet-aware image Transformer that decomposes generation into coarse and fine bands via lossless wavelets. True to its name, the high-frequency bands wait for the signal: staying pure noise until coarse structure has emerged, then joining the flow for joint refinement. Since standard FID discards fine-grained detail through aggressive downsampling, we introduce a more stringent three-axis evaluation protocol to assess quality at native resolution. On ImageNet 512x512, WaiT achieves a pixel-space FID of 1.43 and is Pareto-optimal across all three axes, reducing sampling compute by up to 50%. With our largest 2B model, we set a new state-of-the-art FID of 1.3 for pixel-space models on ImageNet 512 resolution. Our formulation outperforms even the strongest latent-space models on texture fidelity, and scales seamlessly to high-resolution OpenImages and to video generation, achieving a state-of-the-art FVD of 0.84 on Kinetics-600 with no algorithmic modifications.
ER -