Nodes/ComfyUI-H3-Context-Noise/H3 Context Taper Noise (latent)
ComfyUI Node

H3 Context Taper Noise (latent)

Inject tapered gaussian noise into the tail latent steps of an H3 AV context latent, for H3 Motion Context's context_latent input. Audio passes through untouched. Output is for context_latent only -- do not wire it into a VAE decode.

By beijinren·Created 8 days ago·Updated 7 days ago· 3
H3 Context Taper Noise (latent)
  • context_latent
  • context_latent
  • alpha_schedule
tail_frames22
alpha0.45
alpha_end0.10
ramp_steps2
seed0
noise_scalematch_latent_std
Categoryconditioning/minimax

Inputs (7)

NameTypeDefaultDescription
context_latentLATENTThe previous clip's saved AV latent, straight from H3 Motion Context Load Latent.
tail_framesCOMBO22Pixel frames to inject, converted to latent steps. Only these lengths are whole numbers of steps: 5/22/39/56 -> 2/7/12/17 steps. Set this to the same value as Motion Context's context_length.
alphaFLOAT0.450–1Flat injection strength across the pinned steps.
alpha_endFLOAT0.100–1Strength at the last latent step, the one touching the seam.
ramp_stepsINT21–64Latent steps at the end that taper down to alpha_end. Steps are coarse: on a 22-frame window one step is 4 pixel frames, so 2 is roughly the pixel recipe's 3-frame ramp.
seedINT00–18446744073709550000
noise_scaleoptCOMBOmatch_latent_stdmatch_latent_std scales the gaussian to the carried window's own standard deviation, so alpha means the same thing whatever the latent's scale. unit uses raw N(0,1).

Outputs (2)

NameTypeDescription
context_latentLATENT
alpha_scheduleSTRING