Nodes/ComfyUI-MiniMaxH3-FlowDirector/MiniMax H3 Flow Director
ComfyUI Node

MiniMax H3 Flow Director

Visual timeline director for MiniMax H3. Each block on the timeline is generated sequentially as its own flow, automatically passing the decoded last frame to the next block for continuous long videos with low, constant VRAM usage.

By AlonAshken·Created 7 days ago·Updated 7 days ago· 19
MiniMax H3 Flow Director
  • clip
  • vae
  • sampler
  • sigmas
  • model
  • model_ref2va
  • audio_vae
  • noise
  • ref_images
  • first_frame_override
  • model
  • positive
  • latent
  • images
  • audio
  • fps
  • width
  • height
  • length
  • prompt
  • last_image
  • retake_info
timeline_data
global_prompt
start_second0.00
end_second5.00
duration_seconds5.00
start_frame0
end_frame120
duration_frames120
use_custom_audiofalse
use_custom_motiontrue
inpaint_audiotrue
local_prompts
segment_lengths
frame_rate24
display_modeseconds
guide_strength
custom_width0
custom_height0
resize_methodcrop
divisible_by32
img_compression0
override_audiofalse
ref_image_sizematch
shift_video12.00
shift_audio3.00
ref_image_notes
start0.00
end0.00
duration0.00
width0
height0
CategoryMiniMax H3

Inputs (41)

NameTypeDefaultDescription
clipCLIPMiniMax Qwen3-VL text encoder.
vaeVAEminimax_h3_video_vae.
samplerSAMPLERSampler from KSamplerSelect.
sigmasSIGMASSigmas from BasicScheduler.
timeline_dataSTRINGJSON state of the timeline editor (auto-managed).
modeloptMODELfl2va weights (minimax_h3_fl2va_*), used when toolbar is 'Refs OFF'.
model_ref2vaoptMODELref2va weights (minimax_h3_ref2va_*), used when toolbar is 'Refs ON'.
audio_vaeoptVAEminimax_h3_audio_vae for joint audio generation and decoding.
noiseoptNOISENoise from RandomNoise.
global_promptoptSTRINGConditions all blocks: style, atmosphere, characters.
start_secondoptFLOAT0.000–10000
end_secondoptFLOAT5.000–10000
duration_secondsoptFLOAT5.000.1–10000
start_frameoptINT00–1000000
end_frameoptINT1201–1000000
duration_framesoptINT1201–1000000
use_custom_audiooptBOOLEANfalse
use_custom_motionoptBOOLEANtrue
inpaint_audiooptBOOLEANtrue
local_promptsoptSTRING
segment_lengthsoptSTRING
frame_rateoptFLOAT241–240
display_modeoptCOMBOseconds2 options: frames, seconds
guide_strengthoptSTRING
custom_widthoptINT00–8192
custom_heightoptINT00–8192
resize_methodoptCOMBOcrop5 options: maintain aspect ratio, stretch to fit, pad, pad green, crop
divisible_byoptINT321–256
img_compressionoptINT00–100
override_audiooptBOOLEANfalse
ref_image_sizeoptCOMBOmatch2 options: match, max
shift_videooptFLOAT12.000.01–100
shift_audiooptFLOAT3.000.01–100
ref_imagesoptIMAGEExtra reference images for ref2va.
ref_image_notesoptSTRING
startoptFLOAT0.00
endoptFLOAT0.00
durationoptFLOAT0.00
widthoptINT0
heightoptINT0
first_frame_overrideoptIMAGEOptional external image to seed Block 0 (overriding timeline image).

Outputs (12)

NameTypeDescription
modelMODELPatched model output.
positiveCONDITIONINGConditioning output.
latentLATENTLatent output.
imagesIMAGEFull concatenated and decoded video frames (24 fps). Wire directly into VHS Video Combine.
audioAUDIOFull concatenated and decoded stereo audio waveform (44.1 kHz). Wire directly into VHS Video Combine.
fpsFLOATAlways 24.0 — MiniMax H3 native output rate. Wire into VHS Video Combine.
widthINTOutput canvas width.
heightINTOutput canvas height.
lengthINTTotal rendered frame count.
promptSTRINGThe compiled storyboard prompt.
last_imageIMAGELast frame of the final block.
retake_infoSTRINGRetake window info.