Qwen TeleStyle Unload

The Loader builds its pipeline once and keeps it cached, so your next queue doesn't re-read 54GB of weights. That's great right up until it isn't - you switch to a completely different workflow and a bf16 20B model is still sitting on your VRAM, doing nothing, refusing to leave. This is the node that ejects it.

What it does

No inputs, no settings, nothing to configure. Queue it and it moves the cached TeleStyle pipeline off to CPU, drops the references, runs a garbage collection pass, and clears ComfyUI's CUDA cache. It calls the same clear_pipeline_cache() the Loader uses internally when you switch architecture or device - so you never have to care how that works, you just pull the lever.

Its only output is a status string. Wire it into a Show Text node if you want to see "Qwen TeleStyle pipeline cache cleared" printed, though the real evidence is your free-VRAM counter jumping back up.

When to reach for it

Treat it like a del for the pack's resident model. If you're doing a batch of TeleStyle runs back to back, leave the pipeline cached - that's the point. The moment you load a different heavy workflow, drop an Unload node in the queue before anything else. It's not strictly required - restarting ComfyUI does the same thing with more ceremony - but it's the polite, in-graph way to get that memory back.

What it doesn't do

It doesn't delete any of the downloaded weights from disk; those stay in your model folders for next time. It only touches this pack's own pipeline cache - the other models ComfyUI has loaded live their own lives and are unaffected. And there's no "unload every model in the graph" ambition here; it's a single-purpose node, and that's fine.

Installing

Same pack as the Loader and Apply: search ComfyUI-Qwen-TeleStyle in ComfyUI Manager, or cd ComfyUI/custom_nodes && git clone https://github.com/Mister-Link/ComfyUI-Qwen-TeleStyle, then restart. Note it has the same dependency caveat as its siblings - the pack ships no requirements.txt, so a pip install transformers diffusers into ComfyUI's venv may be needed if you see import errors at startup.