MiniMax H3 Scene Data Extract

In the 0.4 days, this pack's scene nodes spat out a dozen separate wires - prompt over here, seed over there, width and height across the canvas. The 0.6 redesign gathered all of that into one typed scene_data bundle, which fixed the graph clutter but created a small new problem: how do you get a single value back out when some other node needs it? MiniMax H3 Scene Data Extract is the answer. It takes the bundle in, you pick a field from a dropdown, and it hands you that one value on its output socket.

The neat part is that the socket's type is dynamic. Pick "Scene number" and the output becomes an INT; pick "Scene prompt" and it's a STRING; pick "Source audio slice" and it becomes an AUDIO socket you can actually wire into an audio consumer. The frontend updates the visible type to match your dropdown selection, so you don't have to guess.

What it can extract

The dropdown covers every value the modern scene bundle carries. The ones you'll actually use, most days:

• Scene prompt / Compiled prompt - the raw scene text versus the prompt after Tagged Ref2VA compiled tags into it. Useful for logging or feeding a preview node.
• Scene number, Scene count, Shot ID - loop bookkeeping when you're driving a display or a conditional.
• Noise seed, Steps, Width, Height, Raw length - anything downstream that needs to mirror the current scene's settings.
• The audio facts: Audio start, Audio duration, Source audio slice.
• The reference diagnostics: Active references, Reference fingerprint, RefMod sources, plus the tagged-scene option values (Reference image size, Reference policy, Semantic anchor size, Semantic anchor mode, Cache for upscale, Conditioning backend, Align audio reference).

That last block is why extract is handy when you're debugging: the bundle that flows through Current Tagged Ref2VA Scene also embeds the options that shaped the scene, so you can surface exactly what a given scene used.

How to use it

Feed scene_data from MiniMax H3 Current Tagged Ref2VA Scene (no other source produces the current bundle format), pick the field, and wire value wherever it needs to go. The pack's advice is to use it sparingly - only pull the facts your workflow genuinely needs. An extract node per scene is fine; ten extract nodes because you wanted every number on the canvas is how a neat redesign turns back into spaghetti.

How to install

It's part of the same pack:

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 via Manager by searching "MiniMax H3 Context Loop").

Common issues

If the node errors with "scene_data does not contain the selected value" or a version complaint, you've wired in a bundle from an older node generation - the 0.4 Current Shot / Tagged Ref2VA nodes don't produce the modern scene_data format, so they can't feed this. And if your output socket looks wrong, re-check the dropdown; the socket type follows the field you selected, so a field change after wiring can leave a dangling connection that now disagrees with its socket.