Simple Trellis2 Clear Cache

This is the smallest node in the pack and the easiest to ignore until the moment you need it. It takes no inputs, runs one function, and returns a single string: "SimpleTrellis2 cache cleared." All it does is unload the TRELLIS.2 pipeline from memory.

Why does that exist? The pack's main node, Simple Trellis2 Image to GLB + Mesh, keeps its loaded pipeline in a cache when keep_model_loaded is on (the default), so your second generation doesn't pay the multi-minute load cost again. That's great. It also means a 4B-parameter model's worth of weights is sitting in VRAM after you're done generating - and TRELLIS.2 is not small.

The Clear Cache node runs exactly the cleanup the main node would: it moves any cached pipeline back to CPU, drops the cache, and empties the CUDA cache (a gc.collect(), torch.cuda.empty_cache(), plus ComfyUI's soft_empty_cache under the hood). The status string is really just a "done" confirmation; there's nothing to wire it into.

When to reach for it

• You're switching model_path directories or toggling low_vram mid-session and want the old pipeline truly gone, not parked in the cache.
• You're done with 3D for now and want the VRAM back for a text-to-image or video job without restarting ComfyUI.
• You changed pipeline parameters and want a clean slate instead of stale state hanging around.

The honest caveat: it clears every cached pipeline (there's usually one), and it only touches memory - nothing on disk, no saved GLB is affected. It's a memory-hygiene tool, not a reset. If generation is actually failing, run the pack's checker (python custom_nodes/ComfyUI-SimpleTrellis2/install.py --check) instead; the cache wasn't your problem.

Install

It ships in the same pack as the main node, so the install is identical:

cd /path/to/ComfyUI/custom_nodes
git clone https://github.com/starsFriday/ComfyUI-SimpleTrellis2.git
cd /path/to/ComfyUI
python custom_nodes/ComfyUI-SimpleTrellis2/install.py

If you already run Image to GLB + Mesh, you already have this node - it's free, found under Trellis2/Simple in the node menu. Fire it whenever you want the explicit unload; it's safe mid-session because the next generation simply reloads the pipeline.