ComfyUI-FDG

Implementation of Guidance in the Frequency Domain Enables High-Fidelity Sampling at Low CFG Scales for ComfyUI.

Thanks to the FDG researchers for making their code public.

FDG improves image quality at low guidance scales and avoids the drawbacks of high CFG scales by design.

Paper: https://arxiv.org/abs/2506.19713

Usage

To use FDG, just place the FDGNode (located in the advanced/model category) in front of the KSampler node in your workflow.

Basically disables cfg values ​​within KSampler.

Testing on SDXL only.

Inputs

• guidance_scale_high: Guidance scale for high-frequency details.
• guidance_scale_low: Guidance scale for low-frequency structures.
• levels: Number of pyramid levels for frequency decomposition. For levels higher than 2, it operates using linear interpolation for each frequency.
• fdg_steps: Number of initial steps where FDG is applied before switching to CFG. Beyond this threshold, the cfg value from the KSampler node takes effect. When cfg equals 1, the guidance_scale_high value is used. If the threshold exceeds the total number of steps, FDG is automatically applied to all steps.

Citation

If you use this implementation in your research, please cite the original paper:

@misc{sadat2025guidance,
    title={Guidance in the Frequency Domain Enables High-Fidelity Sampling at Low CFG Scales},
    author={Seyedmorteza Sadat and Tobias Vontobel and Farnood Salehi and Romann M. Weber},
    year={2025},
    eprint={2506.19713},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}