Nodes/ComfyUi-MpiNodes/Mpi H3 References
ComfyUI Node

Mpi H3 References

MiniMax H3 reference conditioning with every slot exposed at once: 9 images, 3 videos with paired soundtracks, 3 standalone audio. Empty slots are dropped and the rest renumbered, so ONE graph covers every combination instead of a branch per combination. An empty slot means either nothing connected or an Mpi loader with block_if_empty OFF (a 1x1 image / 1-sample waveform) - real black images and real silence are not mistaken for empty. The conditioning is built by core's own MiniMaxH3ReferenceToVideo. Write the prompt against the SLOT numbers - <Picture 1> means ref_image_1 - and the node rewrites them to the ordinals core presents, dropping any tag whose slot is empty. That is what keeps a tag pointing at the same input when a slot ahead of it is left empty, and it is the only way to be right about audio: a reference video's soundtrack consumes an <Audio j>, and whether the file HAS one is unknown until it is decoded. `ref_tags` reports the resulting map.

By MadPonyInteractive·Created 10 months ago·Updated 3 days ago· 2
Mpi H3 References
  • clip
  • vae
  • audio_vae
  • ref_image_1
  • ref_image_2
  • ref_image_3
  • ref_image_4
  • ref_image_5
  • ref_image_6
  • ref_image_7
  • ref_image_8
  • ref_image_9
  • ref_video_1
  • ref_video_audio_1
  • ref_video_2
  • ref_video_audio_2
  • ref_video_3
  • ref_video_audio_3
  • ref_audio_1
  • ref_audio_2
  • ref_audio_3
  • positive
  • latent
  • ref_tags
prompt
width1344
height768
length124
ref_image_sizematch
CategoryMpiNodes/Utils

Inputs (26)

NameTypeDefaultDescription
clipCLIP
vaeVAE
audio_vaeVAE
promptSTRINGAddress references by their SLOT number here: <Picture 1> is whatever is wired into ref_image_1, <Audio 2> into ref_audio_2. The node rewrites them to the ordinals core presents (see ref_tags) - so a tag keeps meaning the same input no matter which other slots are empty. A tag naming an empty slot is removed.
widthINT134432–16384
heightINT76832–16384
lengthINT1245–3600Frame count at 24 fps (124 = ~5 s, trained range ~124-362). Feed it from MpiH3Length.
ref_image_sizeCOMBOmatchHow each reference is scaled (aspect kept, never upscaled, no crop). 'match' fits it to the OUTPUT's pixel area. 'max' uses a 2048 short edge and is best for identity - use it for character sheets, where 'match' squashes the individual views past readability. 'max' is several times slower: reference tokens ride through every sampling step.
ref_image_1optIMAGEReference image. Leave unconnected, or feed a loader with block_if_empty OFF, to skip it.
ref_image_2optIMAGEReference image. Leave unconnected, or feed a loader with block_if_empty OFF, to skip it.
ref_image_3optIMAGEReference image. Leave unconnected, or feed a loader with block_if_empty OFF, to skip it.
ref_image_4optIMAGEReference image. Leave unconnected, or feed a loader with block_if_empty OFF, to skip it.
ref_image_5optIMAGEReference image. Leave unconnected, or feed a loader with block_if_empty OFF, to skip it.
ref_image_6optIMAGEReference image. Leave unconnected, or feed a loader with block_if_empty OFF, to skip it.
ref_image_7optIMAGEReference image. Leave unconnected, or feed a loader with block_if_empty OFF, to skip it.
ref_image_8optIMAGEReference image. Leave unconnected, or feed a loader with block_if_empty OFF, to skip it.
ref_image_9optIMAGEReference image. Leave unconnected, or feed a loader with block_if_empty OFF, to skip it.
ref_video_1optIMAGEReference video frames at 24 fps (2-15 s).
ref_video_audio_1optAUDIOSoundtrack for ref_video_1. Pairing is by this number, so wire it from the same loader.
ref_video_2optIMAGEReference video frames at 24 fps (2-15 s).
ref_video_audio_2optAUDIOSoundtrack for ref_video_2. Pairing is by this number, so wire it from the same loader.
ref_video_3optIMAGEReference video frames at 24 fps (2-15 s).
ref_video_audio_3optAUDIOSoundtrack for ref_video_3. Pairing is by this number, so wire it from the same loader.
ref_audio_1optAUDIOStandalone reference audio.
ref_audio_2optAUDIOStandalone reference audio.
ref_audio_3optAUDIOStandalone reference audio.

Outputs (3)

NameTypeDescription
positiveCONDITIONING
latentLATENT
ref_tagsSTRING