ComfyUI Node

VIG H3 Cutter

A video editor's console for MiniMax H3: several clips on one timeline, each with its own mode, prompt, seed and length, rendered in one pass and re-rendered one at a time. Drag a boundary to move time between two clips; edit a prompt and only that clip and the ones that open on it are rendered again.

By Weloyo·Created about a month ago·Updated 20 days ago· 1
VIG H3 Cutter
  • model
  • ref_model
  • clip
  • vae
  • audio_vae
  • latent_image
  • storyboard
  • images
  • audio
  • frame_count
  • board
  • report
◄board►
◄width832►
◄height480►
◄steps20►
◄sampler_nameres_multistep►
◄schedulersimple►
◄jointrue►
CategoryVIG/MiniMax H3

Inputs (14)

NameTypeDefaultDescription
modelMODELThe H3 base diffusion model, after any patches. Samples the four base modes: t2va, i2va, fl2va, l2va (the fl2va checkpoint covers them all).
boardSTRINGThe timeline, as JSON. Edited through the panel above.
widthINT83232–8192—
heightINT48032–8192—
stepsINT201–200—
sampler_nameCOMBOres_multistep44 options: euler, euler_cfg_pp, euler_ancestral, euler_ancestral_cfg_pp, heun, heunpp2, +38
schedulerCOMBOsimple9 options: simple, sgm_uniform, karras, exponential, ddim_uniform, beta, +3
joinBOOLEANtrueJoining holds every frame of every clip in memory at once -- 3.6 GB a clip at 1344x768. Off gives the clips and nothing else.
ref_modeloptMODELThe H3 ref2va diffusion model. Needed only when a clip is in REF2VA mode; the base modes never touch it.
clipoptCLIPThe minimax CLIP (Qwen3-VL). Required.
vaeoptVAEThe H3 video VAE. Required.
audio_vaeoptVAEThe H3 audio VAE. Required for sound, and for ref2va.
latent_imageoptLATENTSets the canvas: the latent's own proportions override width/height. Wire an Empty Latent Image sized how you want the frame.
storyboardoptVIG_H3_STORYBOARDA whole film from the Film Director. Applied ONCE: the board records which storyboard it was laid out from, so a re-run never puts the plan back over clips you have edited since. Change the script and the new plan applies.

Outputs (5)

NameTypeDescription
imagesIMAGEEvery clip's frames, in order, with the repeated frame at each join removed. Wire to Create Video.
audioAUDIOThe film's soundtrack, in step with the frames.
frame_countINTHow many frames came out.
boardSTRINGThe board as JSON, after the run -- what rendered and what it is cached as.
reportSTRINGWhat ran, what was reused, and every warning the clips raised.