Nodes/RunningHub MiniMax H3/RunningHub MiniMax H3 Ref2VA Video Reference (Legacy)
ComfyUI Node

RunningHub MiniMax H3 Ref2VA Video Reference (Legacy)

Feed MiniMax H3 a video reference, audio track included if you want it

By RH-RunningHub·Created 29 days ago·Updated 18 days ago· 1
RunningHub MiniMax H3 Ref2VA Video Reference (Legacy)
  • video
  • references
  • references
reference_typevideo

The old Ref2VA chain in this pack was built from little reference nodes that each appended one piece of media to a growing list. RHMiniMaxH3Ref2VAVideoReference was the video one - and it had a neat trick: it could add the clip's audio as a reference too, not just the picture.

How it works

Two required inputs:

  • video - a ComfyUI VIDEO (a loaded clip, not frames).
  • reference_type - video or video_audio. Choose video_audio and the node probes the clip for an audio track and, if present, bundles the sound in as an audio reference along with the picture. Choose video and it's visuals only.

One optional input, references (type MINIMAX_H3_REFERENCES) - the list you're appending to. Leave it disconnected and the node starts a fresh list. The single output is that references list, ready to chain into the next reference node or into RHMiniMaxH3Ref2VATarget.

To make the video_audio choice work, the node probes the file with ffmpeg/ffprobe - and it's strict: if you pick video_audio and the clip has no audio track, it raises rather than silently dropping the sound. It also records the video's display dimensions and audio duration into the reference, because the target-resolution logic uses those to decide the output shape.

Why you won't use it

Deprecated, same story as the rest of the legacy set: kept registered so old workflows load, hidden from search. RHMiniMaxH3RefGen replaced the whole append-a-node-at-a-time dance with autogrow ports - and it kept the behavior, so a reference video with an audio track automatically doubles as an audio reference there. If you're on a modern workflow, the only thing worth carrying over is the mental model: a video_audio reference is how you get motion and sound into the generation from one clip.

The one trap that survives

The ffmpeg/ffprobe requirement didn't go away with the legacy node. H3's reference paths need the system binaries on PATH - the pack's requirement list calls them out explicitly. If an old workflow fails right at this node with a probing error, that's the usual cause, and it'll hit RHMiniMaxH3RefGen the same way.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/RH-RunningHub/ComfyUI-RH-MiniMax-H3.git
pip install -r ComfyUI-RH-MiniMax-H3/requirements.txt

Restart, then the ~95 GiB INT8 ConvRot bundle into models/MiniMax-H3-INT8-CONVROT/ - and apt install ffmpeg (or your platform's equivalent) if probing is missing. For new work, honestly, just open RHMiniMaxH3RefGen and drag a video into the autogrow port.

CategoryRunningHub/MiniMax H3/ref2va

Inputs (3)

NameTypeDefaultDescription
videoVIDEO
reference_typeCOMBOvideo2 options: video, video_audio
referencesoptMINIMAX_H3_REFERENCES

Outputs (1)

NameTypeDescription
referencesMINIMAX_H3_REFERENCES