Nodes/AUN ComfyUI Nodes/Inputs Diffusers Refine Basic
ComfyUI Node

Inputs Diffusers Refine Basic

A lightweight diffusion-model setup node that loads a standalone UNet with explicit CLIP and VAE files, prepares common sampler settings, creates an empty latent batch, and optionally provides a separate refinement diffusion model. Right-click → "Collapse Connections" or double-click to hide output labels and converge connection lines.

By loz2754·Created 7 months ago·Updated about a month ago· 4
Inputs Diffusers Refine Basic
    • MODEL
    • CLIP
    • VAE
    • MODEL REFINE
    • model name
    • model name refine
    • sampler
    • scheduler
    • cfg
    • steps
    • latent
    • width
    • height
    • seed
    • batch size
    • speed lora ratio
    diffusion_name
    refine_diffusion_nameNone
    clip_name
    speed_lorafalse
    speed_lora_modelNone
    speed_lora_strength1.00
    speed_lora_full_bothfalse
    speed_lora_ratio1.00
    clip_typeAce
    vae_name
    sampler
    scheduler
    cfg2.0
    steps10
    width720
    height720
    aspect_ratio
    aspect_modeOriginal
    batch_size1
    seed0
    megapixels1.0
    multiple8
    CategoryAUN Nodes/Loaders+Inputs

    Inputs (22)

    NameTypeDefaultDescription
    diffusion_nameCOMBOPrimary diffusion-model file (UNet only). Matches the comfy-core diffusion model list.
    refine_diffusion_nameCOMBONoneOptional refinement diffusion-model file. Select 'None' to reuse the main model.
    clip_nameCOMBOCLIP file shared by the main and refine diffusion models.
    speed_loraBOOLEANfalseEnable SpeedLoRA when loading models.
    speed_lora_modelCOMBONoneSpeedLoRA file to apply after loading the models.
    speed_lora_strengthFLOAT1.000–3Strength multiplier when applying the SpeedLoRA weights.
    speed_lora_full_bothBOOLEANfalseApply the full SpeedLoRA strength to both the main and refine models.
    speed_lora_ratioFLOAT1.000–1Share of the SpeedLoRA strength applied to the main model. The refine model receives the remaining share.
    clip_typeCOMBOAceClip architecture to use when loading a diffusion model.
    vae_nameCOMBOVAE checkpoint shared by the main and refine diffusion models.
    samplerCOMBOSampling algorithm.
    schedulerCOMBONoise schedule.
    cfgFLOAT2.0-2–100Classifier-Free Guidance scale.
    stepsINT101–10000Sampling steps.
    widthINT72064–8192Image width when aspect_ratio='custom'.
    heightINT72064–8192Image height when aspect_ratio='custom'.
    aspect_ratioCOMBOPreset aspect ratio that overrides width/height.
    aspect_modeCOMBOOriginalRandom swaps dimensions 50% of the time, Swap always flips width/height.
    batch_sizeINT11–64Latent batch size.
    seedINT00–18446744073709550000Base seed.
    megapixelsFLOAT1.00.1–16Target total megapixels used when a ratio is selected.
    multipleINT88–128Nearest multiple to round computed resolution to. Used with ratio.

    Outputs (16)

    NameTypeDescription
    MODELMODEL
    CLIPCLIP
    VAEVAE
    MODEL REFINEMODEL
    model nameSTRING
    model name refineSTRING
    sampler*
    scheduler*
    cfgFLOAT
    stepsINT
    latentLATENT
    widthINT
    heightINT
    seedINT
    batch sizeINT
    speed lora ratioFLOAT