Colorful-Noise

Training-Free Low-Frequency Noise Manipulation for Color-Based Conditional Image Generation

SIGGRAPH 2026  |  Nadav Z. Cohen, Ofir Abramovich, Ariel Shamir  |  Project Page  |  arXiv 2605.00548  |  Video

Give the model a color image (a rough color layout, sketch-like coloring, or photo) and a text prompt. Colorful-Noise mixes the low-frequency band of the SDXL initial noise latent (via a radial FFT swap) with the VAE-encoded latents of your color image, biasing generation toward your color layout — no training or fine-tuning required. This Space runs the FFT variant on stock stabilityai/stable-diffusion-xl-base-1.0.

Citation

@misc{cohen2026colorfulnoise,
      title={Colorful-Noise: Training-Free Low-Frequency Noise Manipulation for Color-Based Conditional Image Generation},
      author={Nadav Z. Cohen and Ofir Abramovich and Ariel Shamir},
      year={2026},
      eprint={2605.00548},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      doi={https://doi.org/10.1145/3799902.3811104},
      url={https://arxiv.org/abs/2605.00548},
}