Hunyuan Cache Status V2

Every model on a big-GPU workflow has a "where is it right now?" problem, and Hunyuan 3 makes it worse than most because it can live in three places: fully on the GPU, parked in CPU RAM (after a soft unload), or gone entirely. Hunyuan Cache Status V2 (class HunyuanCacheStatusV2) is the pack's V2-family answer to that question - a zero-input node that just reports what the model cache looks like.

This comes from the same V2 rewrite that gave the pack its unified cache handling (keep_loaded / soft_unload_to_cpu / full_unload post-actions on every generate node). If you've been bouncing between generate nodes and unloaders and you're not sure whether your next run is going to be instant (model still cached) or a two-minute reload, this node tells you before you waste a cycle finding out the hard way.

How it works

It's a diagnostic, not a processor. It reads the pack's shared V2 cache state - which Hunyuan model object is cached, whether it's on GPU or CPU, and roughly how much memory it's holding - and dumps that to a single status string. There's nothing to configure and nothing to wire in.

The inputs and outputs

There are none on the input side. The one output is:

• status (STRING) - the human-readable cache report. Drop the node anywhere in the graph and read the string on the node itself, or route it into a text display node if you want it visible in a saved workflow.

Because it has no inputs, it runs once per graph execution no matter where you place it. That's the whole point: you get a snapshot of the cache at that point in the queue.

How to install it

It ships in the Comfy_HunyuanImage3 pack, so install the pack once and you get this node plus all the loaders, generators and unloaders:

cd ComfyUI/custom_nodes
git clone https://github.com/EricRollei/Comfy_HunyuanImage3
cd Comfy_HunyuanImage3
pip install -r requirements.txt

Then restart ComfyUI. ComfyUI Manager users can just search HunyuanImage in Install Custom Nodes. After the install, hard-refresh the browser (shift-reload) before hunting for the node - a stale frontend hides new nodes even after a clean server restart.

Common issues & troubleshooting

If you see "node not found," the pack didn't load - check the console for the import error and confirm pip install -r requirements.txt actually ran inside the cloned folder. Missing dependencies are the #1 reason these nodes silently vanish.

The other thing to know: this reports the pack's own cache. It won't show you Flux or SAM2 weights sitting in VRAM - that's what Hunyuan 3 Force Unload (Nuclear) with clear_comfy_cache is for, or Hunyuan 3 GPU Info if you want the raw CUDA numbers. For a pure "how much VRAM is my whole system using," reach for those instead; for "is Hunyuan loaded and where," this is the node.