Simple Pixal3D Clear Cache

Every heavy node keeps its models resident between runs, and SimplePixal3D is one of the heavier ones. Simple Pixal3D Clear Cache is the escape hatch: a no-input utility node whose entire job is to unload the Pixal3D pipeline and empty the node's internal cache. Run it when you want your VRAM back without restarting ComfyUI.

Why this exists: the main Simple Pixal3D Image to GLB + Mesh node has keep_model_loaded on by default, so after a generation the whole pipeline - the Pixal3D model plus the DINOv3, RMBG, and MoGe helpers - stays on the GPU. That's exactly why your second run is so much faster than your first. The tradeoff is that those models sit there eating VRAM while you go do something else, and a 24 GB model stack doesn't share politely.

What it actually does

Drop the node in, run the graph, and it calls the pack's _clear_pipeline_cache(): every cached pipeline is moved back to CPU, the cache dict is emptied, and ComfyUI's own memory cache is flushed. Next time you run the generation node it does a full clean load. There's also a real reason to clear even if you're not juggling VRAM: the pipeline cache is keyed by model path, low_vram, and naf_mode, so if you switch those mid-session you can end up with stale state. A clear forces a fresh, correct reload.

What it looks like

About as simple as a node gets. No inputs at all, one output - status (STRING) - and it's an output node, so it fires when the graph runs. It returns a status string ("Simple Pixal3D cache cleared.") that you can wire somewhere or ignore entirely; the point is the side effect, not the value. There's nothing to configure, nothing to get wrong.

Install

Same as the rest of the pack - this node ships in the same repo, so you don't install anything extra. Clone into custom_nodes, run install.py with your ComfyUI's Python, and put the models in models/pixal3d (the Image to GLB article in this pack has the full commands and the pinned CUDA-wheel gotchas). The Clear Cache node is part of the same two-node suite, so if the main node works, this one's already loaded.

When to reach for it

Honestly, only when you need it. If you're batching generations and the second run's speed matters, keep keep_model_loaded on and leave this node out of the graph entirely. If you're bouncing between Pixal3D and another heavy workflow, or you're on a tight-VRAM card where every megabyte counts, it's the difference between a one-node fix and a full ComfyUI restart. Keep it in your back pocket for the end of a big workflow - the day your next job fails to allocate VRAM, you'll be glad it's there.