H3 Capability Probe (alpha)
Ask the running ComfyUI what it can actually do for H3
- model
- model
- report
Every so often a node exists not to make your render better but to make your diagnosis faster. H3 Capability Probe is that node for the modern MiniMax-H3 ComfyUI ecosystem, where half the interesting features arrived as core pull requests and which ones you have depends entirely on how new your ComfyUI install is.
The problem it solves is real and fiddly. H3 features in the running core - per-token masks, clip/audio guides, the tokenizer special-token fix, the condition noise-augmentation dials, the H3 TAE - landed in specific ComfyUI PRs (the node cites #15375, #15439, #15808). Whether a workflow technique works on your box depends on whether your ComfyUI build contains that PR. Guessing wrong means building a graph around a feature your core silently ignores, and the failure is subtle: nothing errors, the feature just doesn't do anything.
The probe reads the installed ComfyUI source and tells you, in one report, exactly which of those capabilities are present. That alone is worth the install if you're doing serious H3 work.
The second thing it does: patch collisions
H3 workflows lean hard on model patching - packs replace double blocks, patch _forward, wrap the final layer, or rewrite core classes. Some of those collide: this pack's own H3 DyRoPE and H3 Streamed Blocks both own the double-block replacement slot, and ComfyUI keeps only one per block. If you've stacked packs (h3-motion-lab, MMH3Tools, Kijai's stuff, this one), it's genuinely hard to keep track of who's patching what.
The probe's report lists every pack that has replaced a double block, patched _forward or the final layer, or rewritten a core class - on your actual installed setup. Wire in a model (optional) and it adds the block replacements and object patches already sitting on that specific model, which is even more useful because that's the model about to go into your sampler.
The inputs and outputs
Required: none. Optional: model (a MODEL to inspect for existing patches). Outputs: model (passed through untouched - nothing here changes a render, and it says so in the node description) and report (a STRING with the full diagnostic). It's an output node, so the report also lands in the UI.
Where it earns its keep
Three specific situations: (1) a capability-dependent graph is misbehaving and you need to know if the core even supports the feature; (2) you're about to stack H3 packs and want to know who's already in the double-block slot; (3) you're upgrading ComfyUI and want to know what new H3 features you just unlocked. It's a "read before you render" tool, and for a pipeline as patch-heavy as H3's, that's a real timesaver.
Installing it
Pure-Python pack, no pip deps:
cd ComfyUI/custom_nodes
git clone https://github.com/matlowai/ComfyUI-MAINodes
Restart, look under MAINodes/alpha. It's alpha because it's new, but it's also one of the safest nodes in the pack - it can't alter a render by construction. And as always with H3, mind the community license (US, EU, UK, South Korea excluded) before you build on the model underneath.
Inputs (1)
| Name | Type | Default | Description |
|---|---|---|---|
| modelopt | MODEL | optional: list the block replacements and object patches already on this model |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| model | MODEL | — |
| report | STRING | — |