Nodes/ComfyUI-Dexogen/H3 Sequence Dexogen
ComfyUI Node

H3 Sequence Dexogen

Generates one or more causally dependent MiniMax H3 audio/video chunks from a validated Dexogen storyboard. The MODEL is consumed after all upstream LoRA and sampling patches. Continuation keeps paired video and audio tails in latent space as native frame-zero AV guides, and can author experimental in-chain edits after the discarded replay overlap. It can resume atomic safetensors chunks from ComfyUI's output directory.

By dexogen·Created 5 days ago·Updated about 24 hours ago· 1
H3 Sequence Dexogen
  • model
  • clip
  • video_vae
  • audio_vae
  • sampler
  • sigmas
  • storyboard
  • first_frame
  • last_frame
  • reference_images
  • av_latent
  • sequence_state
  • report
width1344
height768
cfg1.0
seed0
Category🧬 Dexogen/🎬 Video

Inputs (14)

NameTypeDefaultDescription
modelMODEL
clipCLIP
video_vaeVAE
audio_vaeVAE
samplerSAMPLER
sigmasSIGMAS
storyboardDEXOGEN_STORYBOARD
widthINT134432–16384
heightINT76832–16384
cfgFLOAT1.00–100
seedINT00–18446744073709550000
first_frameoptIMAGE
last_frameoptIMAGE
reference_imagesoptIMAGE

Outputs (3)

NameTypeDescription
av_latentLATENTOne native MiniMax H3 NestedTensor containing the joined video and audio latents.
sequence_stateDEXOGEN_H3_SEQUENCECanonical resumable H3 sequence state without filesystem paths or prompt graph data.
reportSTRINGReadable chunk, seed, timing, and cache summary.