Nodes/h3-latent-upscaler/MiniMax H3 Shift Sigmas (video -> audio)
ComfyUI Node

MiniMax H3 Shift Sigmas (video -> audio)

Remaps a video-schedule SIGMAS array onto MiniMax H3's audio schedule, for feeding the audio branch's MiniMaxH3AddNoise on a pass-2 resample. The DiT internally runs audio on its own shift (default 3.0) while the sampler's sigmas are on the video shift (default 12.0, matching the checkpoint's sampling_settings) -- noising audio with the raw video sigma puts far more noise into it than the model expects at that point in the schedule, which reads as audio distortion. Use the same shift_video/shift_audio as any MiniMaxH3SigmaShift node upstream, or leave at defaults if none is used.

By rockerBOO·Created 28 days ago·Updated 28 days ago· 4
MiniMax H3 Shift Sigmas (video -> audio)
  • sigmas
  • SIGMAS
shift_video12.00
shift_audio3.00
Categorymodel/sampling/sigmas/minimax

Inputs (3)

NameTypeDefaultDescription
sigmasSIGMAS
shift_videoFLOAT12.000.01–100
shift_audioFLOAT3.000.01–100

Outputs (1)

NameTypeDescription
SIGMASSIGMAS