QuadrupleCLIPLoader (Dynamic VRAM)

QuadrupleCLIPLoader (Dynamic VRAM) is the most specific tool in the molbal/ComfyUI-GGUF pack, and the honest review is: you'll know if you need it. It loads four text encoders at once, which is a niche even by ComfyUI standards - most models top out at two or three. It exists so that the pack's CLIP loader family has a counterpart to every ComfyUI core loader, and for the handful of architectures whose conditioning genuinely merges four encoders.

Don't reach for this because you think more encoders means better output - it doesn't. Reaching for it because your workflow file references a QuadrupleCLIPLoader and you want the GGUF version to keep VRAM down? That's exactly the case for it.

How it works

It's the same GGUF loader machinery as the rest of the family, just with four file slots: clip_name1 through clip_name4, all from the combined models/clip + models/text_encoders dropdown. Each accepts .gguf or .safetensors, so you can quantize what's big (a T5, say) and keep the small encoders at full precision. Like the triple variant, there's no type selector - it defaults to stable_diffusion and relies on you wiring the right files. The Dynamic VRAM wrapper manages all four through ComfyUI's aimdo memory manager, and a progress bar tracks each file's tensor loading.

The inputs that matter

• clip_name1 … clip_name4 - the four encoder files. That's the whole interface.

Output is CLIP, into your CLIP Text Encode node.

Installing

Standard pack install - ComfyUI-GGUF in the ComfyUI Manager, or:

cd ComfyUI/custom_nodes
git clone https://github.com/molbal/ComfyUI-GGUF

Restart ComfyUI and pip install --upgrade gguf.

Where people get burned

• Four encoders means four files that all have to be right. There's no architecture selector to save you, so a wrong file produces broken conditioning that can be hard to trace.
• Dynamic VRAM needs a supporting ComfyUI - don't launch with --disable-dynamic-vram, or the loader refuses with a clear error.
• The usual pack caveats: _K quants are text-encoder-only, scaled-FP8 won't mix with GGUF in the same load, and ComfyUI v0.27.0+ is the floor for the native INT8 path.
• And if you're not sure you need four encoders, you probably don't - the dual and triple loaders cover the models people actually run.