FreqFlow

The official implementation of "Frequency-Aware Flow Matching for High-Quality Image Generation"

Flow matching models have emerged as a powerful framework for realistic image generation by learning to reverse a corruption process that progressively adds Gaussian noise. However, because noise is injected in the latent domain, its impact on different frequency components is non-uniform. As a result, during inference, flow matching models tend to generate low-frequency components (global structure) in the early stages, while high-frequency components (fine details) emerge only later in the reverse process. Building on this insight, we propose Frequency-Aware Flow Matching (FreqFlow), a novel approach that explicitly incorporates frequency-aware conditioning into the flow matching framework via time-dependent adaptive weighting. We introduce a two-branch architecture: (1) a frequency branch that separately processes low- and high-frequency components to capture global structure and refine textures and edges, and (2) a spatial branch that synthesizes images in the latent domain, guided by the frequency branch's output. By explicitly integrating frequency information into the generation process, FreqFlow ensures that both large-scale coherence and fine-grained details are effectively modeled low-frequency conditioning reinforces global structure, while high-frequency conditioning enhances texture fidelity and detail sharpness. On the class-conditional ImageNet-256 generation benchmark, our method achieves state-of-the-art performance with an FID of 1.38, surpassing the prior diffusion model DiT and flow matching model SiT by 0.79 and 0.58 FID, respectively.

All models are trained on ImageNet.

accelerate launch --multi_gpu --num_processes 32 --num_machines 4 --main_process_ip $ip --machine_rank $rank \
        --main_process_port $port --mixed_precision fp16 train_ldm_discrete.py \
        --config=configs/freqflow.py

accelerate launch --multi_gpu --num_processes 8 --main_process_port $port --mixed_precision fp16 eval.py \
            --config=./configs/freqflow.py \
            --nnet_path=/path/to/nnet_ema.pth \
            --IMGsave_path=/path/to/ \
            --cfg=$cfg --cfg_scale_pow=1.0 --guidance $guidance

If you have any question, feel free to contact Sucheng Ren

@article{ren2025xar,
       title={Frequency-Aware Flow Matching for High-Quality Image Generation}, 
       author={Sucheng, Ren and Qihang, Yu and Ju, He and Xiaohui, Shen and Liang-Chieh, Chen},
       year={2026},
       booktitle = {CVPR}}