Universal Noise Sampler (Nukun):
NukunUniversalNoiseSampler is a ComfyUI sampling node that lets you choose how noise is generated during image creation. It combines a supplied sampling setup with one of the available Nukun noise profiles, including basic noise types and composite profiles. This gives you a way to explore different noise characteristics without rebuilding the rest of your sampling workflow. It follows a SamplerCustomAdvanced-style approach: your workflow supplies the guidance, sampling algorithm, and noise schedule, while this node applies the selected noise behavior and runs sampling. The source establishes how those pieces are connected, but it does not guarantee a particular visual effect from any profile.
Universal Noise Sampler (Nukun) Input Parameters:
guider
A required GUIDER connection that supplies the guidance setup for sampling. Guidance determines how the sampling process uses your configured conditions. Connect a compatible guider prepared elsewhere in your workflow; this node uses it as supplied rather than creating or configuring one. No default or numeric range is specified.
sampler
A required SAMPLER connection that supplies the sampling algorithm. This node uses the connected sampler together with the guider and noise schedule, so changing it can change how the latent is processed. Select and connect the sampler elsewhere in your workflow. No default or numeric range is specified.
sigmas
A required SIGMAS connection containing the noise-level schedule for sampling. This schedule determines the progression of noise levels through the sampling process; the node does not calculate it for you. Supply a schedule appropriate for the model and sampling setup you have chosen. No default or numeric range is specified.
latent_image
A required LATENT connection containing the starting latent, the working image representation that sampling processes before it is decoded into a viewable image. It may come from a latent-creation step or another compatible part of your workflow. Its contents affect the result, and it must be compatible with the connected sampling setup. No default or numeric range is specified.
add_noise
A BOOLEAN setting that controls whether the node creates noise from the selected Nukun profile. The default is true. Set it to false to use zero noise instead; in that case, the selected noise profile and its generation settings do not produce the noise used for this sampling call. The available values are true and false.
noise_seed
An INT seed used when generating noise from the selected profile. The default is 0, the minimum is 0, and the maximum is 18446744073709551615. Keeping the same seed and other settings helps you compare profile choices consistently; changing it requests a different seeded noise pattern. The control is marked for after-generation seed adjustment. The seed is also passed to the sampling routine and returned as an output.
noise_device
Selects where noise generation is requested to run. The options are auto, cpu, and cuda; the default is auto. According to the node tooltip, auto uses the CPU for ComfyUI-core-like reproducibility, while cuda falls back to the CPU when CUDA is unavailable. Choose cpu when you explicitly want CPU noise generation, or cuda when you want to request it on an available CUDA device. This setting concerns noise generation, not necessarily where every other part of the workflow runs.
noise_profile
Selects the basic noise type or composite Nukun profile used when add_noise is true. The default is gaussian. The available options are gaussian, uniform, laplacian, pink, brown, blue, violet, pyramid, perlin, studentt, white, grey, velvet, green_test, highres_pyramid, pyramid_discount5, pyramid_mix, rainbow_mild, rainbow_intense, wavelet, illustrious_balanced, illustrious_texture, illustrious_composition, illustrious_wild, pony_v7_stage1_gaussian, pony_v7_stage2_violet, pony_v7_balanced, pony_v7_soft, and pony_v7_graphic. Selecting a different option changes the profile requested from the noise generator; the supplied source does not establish a guaranteed visual result for each option.
noise_strength
A FLOAT multiplier applied to the generated noise. The default is 1.0, the minimum is 0.0, the maximum is 5.0, and the adjustment step is 0.01. Higher settings request stronger noise, while lower settings reduce it; 0.0 reduces the generated noise to zero. The tooltip notes that gaussian with auto at 1.0 keeps ComfyUI-core-like behavior. This setting matters when add_noise is true.
detail_bias
A FLOAT control that affects composite noise profiles only. The default is 0.35, the minimum is 0.0, the maximum is 1.0, and the adjustment step is 0.01. Lower values emphasize larger forms in the composite noise, while higher values emphasize details. It does not provide a general detail control for basic profiles, and it has no generated noise to adjust when add_noise is false.
preview_method
Selects the latent preview behavior for this node. The options are default, latent2rgb, taesd, and none; the default is default. Use default to follow ComfyUI’s preview setting, latent2rgb for the lightweight preview described by the tooltip, or none to disable previews. taesd is also an available choice. This setting concerns previews during sampling rather than being a substitute for decoding the final latent into an image.
Universal Noise Sampler (Nukun) Output Parameters:
output
A LATENT containing the sampling result returned by the sampling routine. This is the main result to pass to the next compatible latent-processing step, such as a decoder when you want to view an image. It is still a latent, not a finished image file or an image preview.
denoised_output
A second LATENT returned separately by the sampling routine and named for its denoised result. Use it when a downstream workflow specifically calls for that result, rather than assuming it is interchangeable with output. The supplied source shows that both are returned but does not show the internal computation or guarantee how visibly different they will be.
seed
An INT returning the noise_seed value passed into sampling. Keep it alongside a result to record which seed the node used. It is a seed value, not a measurement of image quality or noise strength.
Universal Noise Sampler (Nukun) Usage Tips:
- Start with
gaussian,noise_deviceset toauto, andnoise_strengthset to1.0when you want the ComfyUI-core-like noise behavior described by the node tooltip. Keep the guider, sampler, sigmas, and latent unchanged while comparing profiles so fewer variables change at once. - Keep
noise_seedfixed while testing differentnoise_profilechoices. Once you find a setting you prefer, change the seed to explore different seeded noise patterns with that configuration. - Adjust
detail_biasonly when testing a composite profile. Move it lower to emphasize larger forms or higher to emphasize details; changing it for a basic profile will not exercise the effect described by the interface. - Set
add_noisetofalsewhen your workflow calls for sampling with zero added noise. Do not expect changes tonoise_profileornoise_strengthto affect generated noise while this switch is off. - Use
latent2rgbwhen you want the lightweight preview identified by the tooltip, ornonewhen you do not need sampling previews. Neither choice changes the type of the returned latent outputs. - Prepare the guider, sampler, and sigmas upstream. This node expects those connections; the separate Nukun Universal KSampler shown in the context is the variant that builds a guider, sampler, and schedule internally.
Universal Noise Sampler (Nukun) Common Errors and Solutions:
Missing required input: guider
- Explanation: The node needs a
GUIDERconnection to direct sampling. This is an illustrative diagnostic; the exact message shown by ComfyUI may differ. - Solution: Connect a compatible guider and make sure its own required model and conditioning setup is complete.
Missing required input: sampler
- Explanation: No
SAMPLERhas been supplied, so the node lacks a sampling algorithm. This is an illustrative diagnostic; the exact message may differ. - Solution: Connect a compatible sampler output to
sampler.
Missing required input: sigmas
- Explanation: The node has no
SIGMASschedule to tell sampling how noise levels should progress. This is an illustrative diagnostic; the exact message may differ. - Solution: Connect a compatible sigma schedule prepared for your sampling setup.
Missing required input: latent_image
- Explanation: Sampling has no starting
LATENTto process. This is an illustrative diagnostic; the exact message may differ. - Solution: Connect a compatible latent source to
latent_image.
Input type mismatch
- Explanation: A connection may provide a different ComfyUI type from the one this input requires. The supplied source does not provide an exact error string.
- Solution: Check the socket labels and connect
GUIDERtoguider,SAMPLERtosampler,SIGMAStosigmas, andLATENTtolatent_image.
Invalid value for noise_seed, noise_strength, or detail_bias
- Explanation: A value outside the node’s declared limits may fail input validation. The supplied source does not provide an exact error string. The limits are
0through18446744073709551615fornoise_seed,0.0through5.0fornoise_strength, and0.0through1.0fordetail_bias. - Solution: Bring the value back within its listed range, particularly when values are supplied programmatically or loaded from an older workflow.
Invalid choice for noise_device, noise_profile, or preview_method
- Explanation: A saved or programmatically supplied choice may not match the options exposed by this node. The supplied source does not provide an exact error string.
- Solution: Re-select the value from the node’s current dropdown. For
noise_device, chooseauto,cpu, orcuda; forpreview_method, choosedefault,latent2rgb,taesd, ornone.
Sampling fails with connected inputs
- Explanation: The node passes the connected guider, sampler, sigmas, and latent into a shared sampling routine. The supplied source does not identify a specific exception for incompatible upstream components or establish which component caused a runtime failure.
- Solution: Check that each upstream node runs with the same intended model and workflow setup. Test with a known-working guider, sampler, sigma schedule, and latent, then change one component at a time to isolate the problem.
No visible preview during sampling
- Explanation:
preview_methodmay be set tonone, ordefaultmay be following a ComfyUI preview setting that does not show the preview you expect. This is a troubleshooting symptom, not a source-defined error message. - Solution: Select
latent2rgbfor the node’s lightweight preview option, or check your ComfyUI preview configuration when usingdefault.
