VOID PQ5 Unload Cache

VOID's PQ5 pipeline caches aggressively. Every VOID PQ5 Load Model bundle - pipeline, VAE, generator, config - stays in memory for the rest of the session, keyed by file paths, so you can re-run a workflow without paying the multi-minute load cost again. That's great until it isn't: load pass 1, load pass 2, iterate a few times, and suddenly you've got several 5B-class video pipelines parked in RAM. This node is the reset button. It clears the whole cache and frees the memory.

It's the rare node with literally no inputs. You drop it in your graph, run it, and it does one thing. The output is a status string ("PQ5 cache cleared") so you have visible confirmation, and ComfyUI's normal execution model means you can even trigger it as part of a workflow rather than restarting ComfyUI.

How it works

It calls the same cache-clear routine the pack uses internally: for every cached pipeline it moves the pipeline and VAE back to CPU, empties the pipeline cache dict, then runs torch.cuda.empty_cache() so the GPU actually hands the freed memory back. The model stays on disk - next time you run a VOID PQ5 Load Model it rebuilds from scratch, but it will not re-download anything. The cost is a cold start on the next load: that multi-minute build you were avoiding is back.

When to reach for it

• Switching between the two-pass checkpoints a lot - each distinct model bundle you load gets cached, and they accumulate fast. Unload between major iterations.
• Out-of-memory on a later node - if a run that used to fit suddenly OOMs, check how many pipeline bundles you've loaded this session. Unload and retry.
• Before a long batch of something else - the VOID pipelines are big; free them before running a different heavy workflow in the same ComfyUI session, rather than after you hit the wall.

The honest trade-off: if you unload before every single run, you've thrown away the whole point of the cache. This node exists for when the cache is costing you, not as a habit. Use it like the memory-pressure valve it is, not after every queue.

Install

Same as the pack: ComfyUI Manager → search "ComfyUI-NetflixVoid", or git clone https://github.com/jaskirat05/ComfyUI-NetflixVoid into custom_nodes, then restart. No dependencies beyond the pack, no downloads, runs instantly. It's the simplest node in the pack - and one of the more useful ones to have around, because memory is the thing that actually kills video workflows.