ComfyUI-CNS-Sampler-CHENGOU Injects Colored Noise For Sharper AI Images

The ComfyUI-CNS-Sampler-CHENGOU is a new custom node that brings Colored Noise Sampling to ComfyUI, improving image generation by routing noise energy to the frequency bands that need it most. Unlike standard samplers that add uniform white noise at every step, it dynamically shifts noise to underbuilt details, boosting sharpness without extra steps. This results in the same model producing noticeably more coherent images with lower FID scores.

Developer IIs-fanta ported the recent Colored Noise Diffusion Sampling into this plug-and-play sampler for FLUX and other diffusion models. The node fills the gap for creators who want to exploit spectral bias in their workflows without complex manual tuning. By making the technique accessible inside the ComfyUI node graph, users can easily swap it in for their current sampler.

Smarter noise allocation across frequency bands

This tool is for ComfyUI users running local diffusion models on consumer GPUs who want better image detail without adding more steps. Small creative studios and hobbyists will appreciate the easy installation and clear preset parameters for FLUX. Privacy-focused professionals benefit because the improvement is fully local and requires no external API calls.

Early days and known limitations

The node is based on a very recent paper, so the developer notes that it’s early-stage and accepts bug reports for edge cases. It includes a built-in approximation of the gamma matrix, meaning no extra files are needed to start, though the official precomputed matrix can be loaded for even better results. As a sampler-level plug-in, it must be used with the SamplerCustomAdvanced node rather than the standard KSampler to function correctly.

“The result is the same model, same step count — with better image quality (lower FID, sharper detail, more coherent structure).” — Source: GitHub