Coupled Initialization
Coupled noise initialization for diverse generation in ComfyUI.
More diversity. Same cost. One node.
Install · Use it · Workflows · How it works · Papers · 中文
</div>Coupled Initialization increases batch diversity without increasing generation cost. Add one node to your existing workflow, with no training or extra dependencies.
The example workflows compare independent and coupled noise with the same prompt, seed and sampling settings.
Independent noise vs. Coupled Initialization
SD1.5 FP16 · 512×512 · K = 2 · seed 538271 · CFG 7 · Euler/Karras · 28 steps · strength 1.
Prompt: A magical library carved inside an ancient giant tree… Coupling produces a different archway composition while preserving the tree library theme.
| Independent Initialization | Coupled Initialization | | :---: | :---: | | <img src="assets/examples/sd15_library_reference.png" alt="Shared first image: a library inside a giant tree" width="205"> <img src="assets/examples/sd15_library_iid_second.png" alt="Second image with independent noise: lanterns among trees" width="205"><br><sub>Shared reference → second candidate</sub> | <img src="assets/examples/sd15_library_reference.png" alt="Shared first image: a library inside a giant tree" width="205"> <img src="assets/examples/sd15_library_coupled_second.png" alt="Second image with coupled noise: an arched tree library" width="205"><br><sub>Shared reference → second candidate</sub> |
The starting noise pair's mean squared distance is 2.01 with independent noise and 3.96 with coupling (1.97×). At K = 2 and strength 1, coupling doubles the expected squared distance.
Install
Link the plugin folder into ComfyUI/custom_nodes:
ln -s /path/to/ComfyUI-CoupledInitialization ComfyUI/custom_nodes/ComfyUI-CoupledInitialization
Restart ComfyUI. There are no Python dependencies.
Use it
Add Coupled Initialization (sampling → custom_sampling → noise), connect its noise output
to SamplerCustomAdvanced → noise in place of RandomNoise, and set the empty latent's
batch_size to K.
| Input | Range | Default | What it does |
| --- | --- | --- | --- |
| seed | 0 to 2⁶⁴−1 | 0 | Seed for the whole batch. Keep it fixed when comparing against IID. |
| strength | 0.0 to 1.0 | 1.0 | 0 reproduces native IID noise exactly; 1 is maximum repulsion. |
| K | ≥ 1 | From latent | K is the latent's batch_size. Start with 2 or 3; K = 1 uses native noise. |
At K = 2, strength = 1 the pair is exactly (ε, −ε), the
antithetic noise of our ICLR 2026 paper,
Antithetic Noise in Diffusion Models.
Example workflows
| Model | Sampling settings | K = 2 | K = 3 | | --- | --- | --- | --- | | SD1.5 | CFG 7, karras, 28 steps, 512×512 | Workflow | | | SDXL | CFG 3, karras, 30 steps | Workflow | Workflow | | FLUX.1-schnell | BasicGuider, simple, 4 steps | Workflow | Workflow |
Each workflow compares Coupled Initialization with native RandomNoise. Both branches share the
same prompt, guider, sampler, sigmas and latent.
The SD1.5 workflow uses Comfy-Org's FP16 checkpoint. Place it in models/checkpoints before loading the workflow.
How it works
Coupled Initialization transforms the batch's initial noise so that any two samples have
correlation ρ = −strength / (K − 1). At strength = 1, this is the most negative correlation
possible for K jointly Gaussian variables. Each sample retains a standard Gaussian marginal;
coupling changes the batch's joint distribution.
- Models: diffusion and flow models initialized with standard Gaussian noise.
- Sampling interface:
SamplerCustomAdvancedand other nodes that consume aNOISEobject. - Latents: image latents shaped
[batch, channels, height, width]. - Precision: half precision batches are reduced in float32; float64 inputs keep double precision.
- Batch indices: use distinct
batch_indexvalues or a fresh empty latent. Setstrength = 0to use native noise.
Paper & citation
Coupled Initialization builds on two papers:
- Couple to Control: Joint Initial Noise Design in Diffusion Models. Jing Jia, Liyue Shen and Guanyang Wang. arXiv, 2026.
- Antithetic Noise in Diffusion Models. Jing Jia, Sifan Liu, Bowen Song, Wei Yuan, Liyue Shen and Guanyang Wang. ICLR 2026.
Please cite both papers if you use Coupled Initialization in research.
<details> <summary>BibTeX</summary>@misc{jia2026couplecontrol,
title={Couple to Control: Joint Initial Noise Design in Diffusion Models},
author={Jing Jia and Liyue Shen and Guanyang Wang},
year={2026},
eprint={2605.11311},
archivePrefix={arXiv},
primaryClass={cs.LG},
url={https://arxiv.org/abs/2605.11311}
}
@inproceedings{jia2026antithetic,
title={Antithetic Noise in Diffusion Models},
author={Jing Jia and Sifan Liu and Bowen Song and Wei Yuan and Liyue Shen and Guanyang Wang},
booktitle={International Conference on Learning Representations},
year={2026},
url={https://proceedings.iclr.cc/paper_files/paper/2026/file/20b6b87ca17792337f414d948af7b0e8-Paper-Conference.pdf}
}