Nodes/ComfyUI-FL-BerniniR/FL Bernini-R Unload
ComfyUI Node

FL Bernini-R Unload

The tiny node that hands your VRAM back

By filliptm·Created 3 months ago·Updated 3 months ago· 0
FL Bernini-R Unload
  • anything
  • status

Bernini-R is a 14B renderer. Loaded, it doesn't politely sit in the corner of your VRAM the way an SDXL checkpoint does - it's the whole party, and it stays invited after your render finishes. If your next workflow is anything lighter, you need FL Bernini-R Unload. It's the smallest node in the pack and arguably the one you'll reach for most.

The mechanism is simple. The Loader keeps a cached pipeline in memory so you can run many prompts against one load - that's the entire point of the loader cache. Unload clears that cache, runs a garbage collect, and empties the CUDA cache. In practice, the model's weights drop out of GPU and system memory and your card's free VRAM goes back up.

It has exactly one input, an anything port (type *), which exists purely so you can wire the end of your Bernini graph into it and force it to run after Generate. Any output works - the node deletes whatever it receives. It returns a status string, and "Bernini-R pipeline cache cleared." shows in the node's text output.

When to use it

Two sensible patterns. The obvious one: render a video, unload, then run your SDXL or Flux workflow in the same session without a manual reload. The tidy one: set unload_after_run on FL Bernini-R Generate instead and let the generate node clean up after itself. Pick one; doing both is harmless but pointless.

The gotcha

This clears the whole Bernini-R cache, not just the pipeline you just used. Load several variants into the same session and they all go. And once unloaded, the Loader's next run has to load the model from scratch - which, on a 117 GB snapshot, is not fast. The right rhythm: load once, run all your Bernini prompts, then unload when you switch models or close the session.

Install

Same as the rest of the pack. ComfyUI Manager users search ComfyUI-FL-BerniniR; otherwise:

cd ComfyUI/custom_nodes
git clone https://github.com/filliptm/ComfyUI-FL-BerniniR.git
cd ComfyUI-FL-BerniniR
pip install -r requirements.txt

Restart ComfyUI after installing.

Troubleshooting

If Unload runs but your VRAM monitor shows no change, check that the Loader actually loaded a pipeline in the first place - no load, nothing to unload. And don't mistake this for file cleanup: it frees memory only. Your rendered .mp4 and .png files in ComfyUI/output/BerniniR stay put. If Generate later complains the pipeline is gone, that's not a bug - you unloaded it. Re-run the Loader.

CategoryFL/BerniniR

Inputs (1)

NameTypeDefaultDescription
anything*

Outputs (1)

NameTypeDescription
statusSTRING