Cache Node

Cache Node writes a latent, an image batch, or a conditioning to disk so you can reuse it later without recomputing it. If you've got an expensive step early in a workflow - a slow encode, a heavy conditioning build, a first-pass generation - you can cache the result once and then keep iterating on everything after it without paying that cost again on every run.

That's the whole pitch: don't recompute what hasn't changed. It's a manual, explicit version of caching - you decide what to stash and when to load it back - which makes it predictable in a way that automatic caching isn't.

How it works

Per the WAS README, Cache Node can cache latents, tensor batches (images), and conditioning to disk. You give it a name, and it writes the data out to a file. Its partner, Load Cache, reads that file back in later - you point Load Cache at the same name and it returns the stored latent/image/conditioning as if the upstream nodes had just produced it.

So the pattern is two nodes: Cache Node to save, Load Cache to restore. Run your expensive graph once with Cache Node capturing the result, then disconnect (or bypass) the expensive part and feed downstream from Load Cache instead. Now you're iterating on the cheap tail of the workflow at full speed.

The inputs and outputs that matter

You wire in whatever you want to stash - a latent, an image batch, or a conditioning - and give the cache a name so you can find it again. The saving happens as a side effect; the real payoff is on the read side, where Load Cache uses that same name to bring the data back. Keep names distinct and descriptive, because the name is your only handle on what's in there.

How to install it

Comes with WAS Node Suite. Install once via ComfyUI Manager (search WAS Node Suite, install, restart) or clone it:

cd ComfyUI/custom_nodes
git clone https://github.com/WASasquatch/was-node-suite-comfyui
pip install -r was-node-suite-comfyui/requirements.txt

then restart. On Windows portable, run the pip step with python_embeded\python.exe -s -m pip install -r ... (or install.bat). No models needed - it just reads and writes files.

Common issues & troubleshooting

Load Cache can't find your data. The name is the link between saving and loading. A typo or a mismatched name and Load Cache has nothing to read. Also: the cache has to have actually been written first - run the graph with Cache Node at least once before expecting Load Cache to return anything.

Stale cache bites you. This is the classic caching trap. If you change something upstream of Cache Node but keep loading the old cached result, you're iterating on outdated data and wondering why your changes do nothing. When you edit anything before the cache, re-run to refresh it. The cache doesn't know your inputs changed - that's on you.

Disk fills up. Cached latents and image batches are real files on disk, and they don't clean themselves up. If you cache a lot, especially big image batches, keep an eye on the folder and clear old caches you no longer need.

The suite won't import. WAS Node Suite is large and unmaintained since late 2023; the recurring failure is an "Import Failed" after a ComfyUI update, from a dependency version clash (opencv being the usual culprit). Reinstall the requirements against your ComfyUI Python and restart. Cache Node itself is straightforward - it's the suite install that occasionally trips.