Nodes/ComfyUI-MiniMaxH3-ChainDirector/MiniMax H3 Chain Director
ComfyUI Node

MiniMax H3 Chain Director

A ComfyUI node in utils/MiniMaxH3 with 34 inputs and 4 outputs.

By luxu1999·Created 3 days ago·Updated about 4 hours ago· 5
MiniMax H3 Chain Director
  • model_r2v
  • model_i2v
  • video_vae
  • audio_vae
  • clip
  • image_0
  • image_1
  • image_2
  • image_3
  • image_4
  • image_5
  • image_6
  • image_7
  • image_8
  • ref_video_0
  • ref_video_1
  • ref_video_2
  • ref_audio_0
  • ref_audio_1
  • ref_audio_2
  • images
  • audio
  • fps
  • frame_count
global_prompt
timeline_prompt
duration_preset30s
split_preset5s per segment (recommended)
resolution_preset0.4MP (480p)
ref_max_size864
auto_anchortrue
steps4
samplerer_sde
schedulersimple
cfg1.00
seed0
shift_video12.00
shift_audio3.00
Categoryutils/MiniMaxH3

Inputs (34)

NameTypeDefaultDescription
model_r2vMODELref2va base model, used for the first r2v segment
model_i2vMODELfl2va base model, used for segment 2+ (i2v handoff)
video_vaeVAE
audio_vaeVAE
clipCLIP
image_0IMAGEReference image 1 → <Picture 1> (scene / main subject)
global_promptSTRINGConstant settings for the whole video: scene/style/character/camera. You may reference images with <Picture N>.
timeline_promptSTRINGRequired. One line per block: 0-5s: action description. Auto-mapped to segments.
duration_presetCOMBO30sOptions: 5/10/15/30/45/60/90/120s
split_presetCOMBO5s per segment (recommended)5/10/15s per segment; max ~15s (362 frames) per segment
resolution_presetCOMBO0.4MP (480p)0.4MP=864x480 (480p) / 0.9MP=1280x736 (720p) / 2.0MP=1920x1088 (1080p)
ref_max_sizeINT864256–2048Max edge length for reference image resizing; usually matches the resolution preset
auto_anchorBOOLEANtrueAuto-append first-frame lock / body-size / reference consistency anchors
stepsINT41–100Diffusion steps INSIDE EACH segment. 4 = turbo LoRA recommended; 8 is sharper but ~2x slower.
samplerCOMBOer_sde44 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
cfgFLOAT1.000–10Recommended 1.0 with the turbo LoRA
seedINT0
shift_videoFLOAT12.00
shift_audioFLOAT3.00
image_1optIMAGEReference image 2 → <Picture 2>
image_2optIMAGEReference image 3 → <Picture 3>
image_3optIMAGEReference image 4 → <Picture 4>
image_4optIMAGEReference image 5 → <Picture 5>
image_5optIMAGEReference image 6 → <Picture 6>
image_6optIMAGEReference image 7 → <Picture 7>
image_7optIMAGEReference image 8 → <Picture 8>
image_8optIMAGEReference image 9 → <Picture 9>
ref_video_0optIMAGEReference video 1 (IMAGE frame batch) → <Video 1>, used by the first r2v segment
ref_video_1optIMAGEReference video 2 (IMAGE frame batch) → <Video 2>
ref_video_2optIMAGEReference video 3 (IMAGE frame batch) → <Video 3>
ref_audio_0optAUDIOReference audio 1 → <Audio 1>, used by the first r2v segment
ref_audio_1optAUDIOReference audio 2 → <Audio 2>
ref_audio_2optAUDIOReference audio 3 → <Audio 3>

Outputs (4)

NameTypeDescription
imagesIMAGE
audioAUDIO
fpsFLOAT
frame_countINT