Nodes/ComfyUI-ZImagePowerNodes/⚗️🔬| Z-Sampler Turbo ^G2.1
ComfyUI Node

⚗️🔬| Z-Sampler Turbo ^G2.1

Experimental new version of Z-Sampler Turbo. It takes a Z-Image Turbo model, an initial latent image, and prompt/conditioning to produce a denoised latent output, which can then be decoded into the final image. ⚠️ Because this node is experimental, its parameters, behavior, or availability may change or be removed entirely without prior notice.

By martin-rizzo·Created 7 months ago·Updated a day ago· 346
⚗️🔬| Z-Sampler Turbo ^G2.1
  • latent_input
  • model
  • positive
  • divider1
  • positive_stg2
  • positive_stg3
  • latent_output
seed1
steps8
initial_bias0.0
latent_scaling
spectral_tilt
turbo_creativityfalse
detailed_refinertrue
new_schedulertrue
Category⚡Z-Image/__experimental

Inputs (14)

NameTypeDefaultDescription
latent_inputLATENTThe initial latent image to be denoised; usually an 'Empty Latent' for text-to-image tasks or an encoded image for image-to-image processing.
modelMODELThe Z-Image Turbo model used for denoising the latent image.
positiveCONDITIONINGThe main prompt/conditioning used to guide the generation process toward the desired content.
seedINT11–18446744073709550000The seed used for the random noise generator, ensuring the same result is produced with the same value.
divider1ZIPN_SEPARATOR
stepsINT82–14Number of iterations to perform during the denoising process.
initial_biasFLOAT0.0-0.5–0.5Custom adjustment for initial noise bias, usually kept at 0.0; Positive values amplify dominant prompt features (e.g., making bright scenes brighter or dark scenes darker), while negative values temper them. Keep in mind it reacts differently to every prompt, it's not a simple brightness control.
latent_scalingCOMBODetermines the resolution scale applied to latents during each sampling stage. 'Max Speed' uses lower internal scaling for faster generation, while 'Max Quality' employs higher scaling to preserve finer details and structural integrity at the cost of processing time.
spectral_tiltCOMBOApplies a frequency-based bias to the noise across sampling stages. Early stages amplify low-frequency noise to establish composition, while later stages shift focus to high-frequency components to sharpen fine details. Each preset defines which stages (1, 2, or 3) receive this spectral modification and the intensity of the adjustment.
turbo_creativityBOOLEANfalseEnables turbo creativity. This scrambles the image to boost diversity in compositions while maintaining the general style and tone color. Be aware that this may lead to hallucinations.
detailed_refinerBOOLEANtrueEnables an alternative refiner using the DPM++ SDE sampler during the final stage. This enhances contrast and sharpness in fine details but increases overall processing time.
new_schedulerBOOLEANtrueEnables the optimized scheduler with an updated set of sigmas for superior general quality. Disabling this switches back to the legacy version, which may still perform better in specific edge cases.
positive_stg2optCONDITIONINGThis input is optional and can remain disconnected. It allows specifying a different prompt/conditioning for the second stage of the denoising process.
positive_stg3optCONDITIONINGThis input is optional and can remain disconnected. It allows specifying a different prompt/conditioning for the third stage of the denoising process.

Outputs (1)

NameTypeDescription
latent_outputLATENTThe resulting denoised latent image, ready for decoding by a VAE or passed to another node for further processing.