Nodes/ComfyUI-MiniMaxH3-Context-Loop/MiniMax H3 Current Tagged Scene Pack (Internal)
ComfyUI Node

MiniMax H3 Current Tagged Scene Pack (Internal)

The (Internal) packer you're only meeting because ComfyUI expanded a node

By ethanfel·Created 30 days ago·Updated about 19 hours ago· 383
MiniMax H3 Current Tagged Scene Pack (Internal)
  • state
  • source_audio_slice
  • refmod_sources
  • options
  • scene_data
clip_index
clip_count
shot_id
prompt
noise_seed
length
steps
width
height
audio_start
audio_duration
status
compiled_prompt
active_references
reference_fingerprint

Be honest: you probably didn't search for this node, you found it. MiniMax H3 Current Tagged Scene Pack is an internal packer whose whole job is to be created on the fly when MiniMax H3 Current Tagged Ref2VA Scene expands. If you're looking at it in your graph, it's because you clicked into the expanded subgraph of that modern node, or you're rebuilding by hand what that node does automatically. Either way, you almost never need to wire it yourself.

Its job is simple and slightly thankless: take about nineteen resolved values from the current scene and squash them into one typed scene_data bundle. Those values are the things Current Shot and Tagged Ref2VA used to expose as separate output sockets in 0.4 - the scene prompt, seed, raw frame length, steps, canvas, shot ID, scene number/count, audio start and duration, the compiled prompt, active references, the full reference fingerprint, the RefMod source list, and the validated options that shaped the scene.

Why bundle instead of wire

Remember what a 0.4 reference graph looked like: Current Shot fanning out a dozen wires (prompt here, seed there, width and height to the sampler), with Tagged Ref2VA contributing more. Every wire is a place to lose track of which value is current. This pack's 0.6 answer is the classic context-bundle move: one typed carrier down one wire, unpacked only where a value is actually needed. The trade-off is real - a bundle hides what's inside until you open it - which is why the pack ships MiniMax H3 Scene Data Extract as the sanctioned way to pull individual values back out.

Inputs and outputs

The node's inputs mirror what it packs: state, clip_index, clip_count, shot_id, prompt, noise_seed, length, steps, width, height, audio_start, audio_duration, source_audio_slice, status, compiled_prompt, active_references, reference_fingerprint, refmod_sources, and options. One output: scene_data, a versioned H3_SCENE_DATA payload that also embeds the tagged-scene options so downstream extractors can read them back with type knowledge.

The versioning matters more than it looks. Because scene_data carries a version tag and every node validates it, a stale bundle from a different pack version fails loudly instead of silently feeding your sampler the wrong seed. And the reference fingerprint packed in here is part of what lets a resumed run reject incompatible generation changes.

How to install

It ships inside the same pack as everything else here:

cd ComfyUI/custom_nodes
git clone https://github.com/seitanism/ComfyUI-H3-Motion-Context-MultiRef.git
git clone https://github.com/ethanfel/ComfyUI-MiniMaxH3-Context-Loop.git

Restart ComfyUI (or install "MiniMax H3 Context Loop" via Manager). No separate dependencies beyond the companion MultiRef repo and a current ComfyUI.

Common issues

The only real failure mode is manual misuse: feeding it a state that isn't a resolved modern Current Shot state, which it rejects with a validation error. If you catch yourself hand-wiring all nineteen inputs, step back - that's what Current Tagged Ref2VA Scene does for you, and doing it by hand just reintroduces the 19-wire mess this pack spent a version removing.

Categoryconditioning/minimax/context_loop/internal

Inputs (19)

NameTypeDefaultDescription
stateH3_CHAIN_STATEResolved Current Shot state.
clip_indexINTResolved one-based scene number.
clip_countINTTotal scene count.
shot_idSTRINGStable current scene ID.
promptSTRINGResolved current scene prompt.
noise_seedINTResolved current scene seed.
lengthINTResolved H3 raw frame length.
stepsINTResolved sampler step count.
widthINTResolved generation width.
heightINTResolved generation height.
audio_startFLOATCurrent scene source-audio start time.
audio_durationFLOATCurrent scene source-audio duration.
source_audio_sliceAUDIOOptional current source-audio reference.
statusSTRINGCurrent Shot diagnostic status.
compiled_promptSTRINGTagged Ref2VA compiled prompt.
active_referencesSTRINGTagged Ref2VA scene reference summary.
reference_fingerprintSTRINGComplete tagged reference lineage.
refmod_sourcesH3_REF_LISTActive visual sources for external RefMod.
optionsH3_TAGGED_SCENE_OPTIONSValidated options used for this scene.

Outputs (1)

NameTypeDescription
scene_dataH3_SCENE_DATAPacked modern Current Shot and Tagged Ref2VA secondary values.