VELVET VICE · H3 PREFLIGHT
A pre-flight checklist you can run before a 33B render commits
There's nothing more deflating than queueing an H3 render and watching it die three minutes in because a model file is missing, RAM is nearly exhausted, or the System Hub is pointing at a backend you don't actually have installed. VelvetViceMiniMaxH3Preflight is the pack's attempt to catch those problems before they cost you GPU time - a read-only monitor that checks the state of the H3 setup and reports back.
Notice the category: VELVET VICE/MiniMax H3/00 Monitor. It's one of a family of monitor nodes (the Render Timer is the other) that don't process data at all. They're instrument panels, deliberately numbered to sit at the front of the workflow like a checklist you glance at before takeoff.
What it does
The preflight has a single input: auto_check_on_load (BOOLEAN, default true). It's an output node with no data outputs - run it and it returns a UI status rather than values for other nodes to consume.
What that status actually checks is the backend and model state of the H3 setup - the README calls it a "MiniMax H3 backend and model preflight." The description adds the modern twist: "the monitor can check the current System Hub selections automatically or on demand." So in the current pack layout it doesn't just scan folders in the abstract; it can look at what your System Hub is currently configured to load and confirm the files behind those selections exist - which is far more useful than a generic "model folder is non-empty" check. AUTO mode runs the check when the workflow loads; you can also trigger it manually when you've just swapped models or backends and want a second opinion before queueing.
How you'd use it
Keep it somewhere visible at the top of the workflow and run it once when you change anything about your setup: new GGUF download, different text encoder, a System Hub backend flip. The check is cheap - it's scanning folders and reading config, not loading weights - so running it on load costs you nothing. If it reports problems, fix them before the queue, not after.
Install
It ships with the pack, so the install is the pack install:
cd ComfyUI/custom_nodes
git clone https://github.com/Velvet-Vice/velvet-vice-minimax-h3
or ComfyUI Manager → velvet-vice-minimax-h3, then restart.
A note on expectations so you're not disappointed: this is a preflight, not a guarantee. It can confirm files exist and the config is coherent; it cannot promise the render won't OOM, because nothing can without actually running it. On an H3-sized workload the honest division of labor is: preflight catches the avoidable errors, System Check monitors RAM during the run, and the render itself is where the remaining surprises surface. And worth repeating whenever anyone mentions local H3: the model's community license excludes the US, EU, UK and South Korea from running the weights, so a "preflight passed" is meaningless if your region never should have loaded them.
Inputs (1)
| Name | Type | Default | Description |
|---|---|---|---|
| auto_check_on_load | BOOLEAN | true | — |
Outputs (0)
No outputs