Nodes/ComfyUI-MiniMax-H3-Extend/MiniMax H3 Video Extend (Backported)
ComfyUI Node

MiniMax H3 Video Extend (Backported)

Continue a prior MiniMax H3 clip from its trailing latent frames (backported, see README.md).

By kat3ri·Created 14 days ago·Updated 14 days ago· 2
MiniMax H3 Video Extend (Backported)
  • clip
  • vae
  • context_latent
  • audio_vae
  • first_frame
  • last_frame
  • ref_images
  • ref_audio
  • positive
  • latent
prompt
length124
context_frames2
pin_last_frametrue
ref_image_sizematch
Categorymodel/conditioning/minimax

Inputs (13)

NameTypeDefaultDescription
clipCLIP
vaeVAE
context_latentLATENTAV latent output from a prior MiniMax H3 generation to continue from
promptSTRING
lengthINT1245–3600New frame count at 24 fps for the continuation only (excludes context_frames)
context_framesINT21–64Trailing latent frames of context_latent carried over as context
pin_last_frameBOOLEANtrueDecode context_latent's true trailing pixel frame and pin it as this call's frame 0. Ignored if first_frame is connected.
audio_vaeoptVAE
first_frameoptIMAGEHard-pin this call's frame 0 to an exact image (e.g. the prior clip's real last output frame) instead of pin_last_frame's decode
last_frameoptIMAGEPin this continuation segment's own final frame to an exact image -- e.g. to land precisely on a known next shot/keyframe instead of leaving the ending fully generated. Not part of the native fork's VideoExtend node (which has no end-anchor at all); added here since it's a natural, low-risk extension of the same PackedLayout mechanism first_frame/context already use.
ref_image_sizeoptCOMBOmatch2 options: match, max
ref_imagesoptIMAGEReference image(s) -- connect a batch (e.g. via ImageBatch) for more than one; each frame becomes its own <Picture i> reference. Not the native node's per-slot Autogrow inputs -- this is a single batched socket.
ref_audiooptAUDIOOne standalone reference audio clip

Outputs (2)

NameTypeDescription
positiveCONDITIONING
latentLATENT