Nodes/comfyui-jjmaden-random-nodes/MiniMax H3 Unified To Video (best-effort)
ComfyUI Node

MiniMax H3 Unified To Video (best-effort)

Best-effort reimplementation of a third-party workflow node with no public source, built from ComfyUI's real native MiniMax H3 building blocks. Combines EmptyMiniMaxH3LatentAV + MiniMaxH3ImageToVideo + MiniMaxH3ReferenceToVideo into one node, tagging first_frame/last_frame and every reference into one <Picture N>/<Video N>/<Audio N> sequence. Reference inputs are named to match the real, already-installed MiniMaxH3ReferenceSplitter's output labels (picture_N/video_N/video_audio_N/audio_N). Wire an H3_REFS `references` bundle straight from MiniMaxH3MediaLoader or MiniMaxH3PromptBuilder instead of using a Reference Splitter -- direct sockets win per-slot when both are wired, the bundle fills in the rest (up to 9 pictures / 3 videos / 3 paired video_audios / 3 standalone audios).

By jjmaden·Created 7 days ago·Updated 3 days ago· 0
MiniMax H3 Unified To Video (best-effort)
  • clip
  • video_vae
  • audio_vae
  • first_frame
  • last_frame
  • references
  • picture_1
  • picture_2
  • picture_3
  • picture_4
  • video_1
  • video_audio_1
  • audio_1
  • audio_2
  • positive
  • av_latent
  • conditioned_prompt
  • media_map_json
  • report
prompt
modeauto
width1344
height768
duration5.0
fps24
ref_image_sizematch
CategoryMiniMax H3/custom

Inputs (21)

NameTypeDefaultDescription
clipCLIP
video_vaeVAE
promptSTRING
modeCOMBOautoauto/hybrid = use whatever is wired. keyframe = ignore ref_* inputs. reference = ignore first_frame/last_frame. text = ignore all media.
widthINT134432–16384
heightINT76832–16384
durationFLOAT5.00.2–150Seconds; converted to a frame count via fps and snapped to H3's grid.
fpsINT241–60
ref_image_sizeCOMBOmatchmatch = scale refs to the generation's pixel area. max = use the 2048px reference pipeline (slower, higher fidelity).
audio_vaeoptVAE
first_frameoptIMAGE
last_frameoptIMAGE
referencesoptH3_REFSWire this straight from MiniMaxH3MediaLoader's or MiniMaxH3PromptBuilder's 'references' output instead of using a Reference Splitter. Fills in whichever picture_N/video_N/video_audio_N/audio_N slot below isn't directly wired (up to 9 pictures / 3 videos / 3 paired video_audios / 3 standalone audios).
picture_1optIMAGE
picture_2optIMAGE
picture_3optIMAGE
picture_4optIMAGE
video_1optIMAGE
video_audio_1optAUDIO
audio_1optAUDIO
audio_2optAUDIO

Outputs (5)

NameTypeDescription
positiveCONDITIONING
av_latentLATENT
conditioned_promptSTRING
media_map_jsonSTRING
reportSTRING