Nodes/comfyui-turbodiffusion/TurboDiffusion I2V Sampler
ComfyUI Node

TurboDiffusion I2V Sampler

Complete TurboDiffusion I2V inference with dual-expert sampling

By anveshane·Created 8 months ago·Updated 6 months ago· 219
TurboDiffusion I2V Sampler
  • high_noise_model
  • low_noise_model
  • conditioning
  • vae
  • image
  • frames
num_frames77
num_steps4
resolution480
aspect_ratio16:9
boundary0.9
sigma_max200.00
seed0
use_odefalse
width480
height480
Categoryvideo/turbodiffusion

Inputs (15)

NameTypeDefaultDescription
high_noise_modelMODELHigh noise expert model from TurboWanModelLoader
low_noise_modelMODELLow noise expert model from TurboWanModelLoader
conditioningCONDITIONINGText conditioning from CLIPTextEncode
vaeVAEWan2.1 VAE from VAELoader
imageIMAGEStarting image for I2V generation
num_framesINT779–241Number of frames to generate (must be 8n+1)
num_stepsCOMBO4Number of sampling steps (1-4 for distilled model)
resolutionCOMBO480Base resolution or custom
aspect_ratioCOMBO16:9Aspect ratio
boundaryFLOAT0.90–1Timestep boundary for switching from high to low noise model
sigma_maxFLOAT200.001–1000Initial sigma for rCM sampling
seedINT00–18446744073709550000Random seed
use_odeBOOLEANfalseUse ODE sampling (sharper but less robust)
widthINT48064–2048Custom width (only used when resolution is 'custom')
heightINT48064–2048Custom height (only used when resolution is 'custom')

Outputs (1)

NameTypeDescription
framesIMAGE