Smart Quad CLIP Loader MXD

Let's be straight with you: the Smart Quad CLIP Loader MXD is the niche one. Four text encoders is not a setup most models need, which is why ComfyUI has no core quad loader at all - this pack added one because it could, and the README itself files it under "advanced/experimental workflows." If you've got a legit four-encoder model (some research checkpoints, custom multi-modal experiments), this is the only clean way to load all four through one node.

What it actually is: the family's SmartCLIPLoaderBase with four clip_name slots, one type dropdown, and the same per-slot format independence as the dual and triple variants. Each slot takes a .gguf or safetensors file from models/text_encoders independently, so a quantized encoder can sit next to three full-precision ones - useful if the four-encoder stack is what's blowing past your VRAM and you only want to shave the biggest piece.

How it works

No new mechanism - it's the same machinery as the dual and triple loaders, itself a fork of city96/ComfyUI-GGUF's loader code. The dropdown merges your text encoders folder with the pack's "clip_gguf" key, and each file loads by extension: GGUFs through gguf_clip_loader with GGML ops, safetensors through the standard torch loader, then everything joins via load_text_encoder_state_dicts into one CLIP. Tokenizers for GGUF T5/Qwen-style encoders are rebuilt from file metadata, which is what sentencepiece and protobuf in the requirements are for.

The type dropdown is the same 25-entry list the core CLIPLoader uses (sd3, stable_cascade, ltxv, cosmos, wan, hidream, chroma, and friends). For a four-encoder model that doesn't match a listed type, the string maps against ComfyUI's CLIPType enum and silently falls back to stable_diffusion on a miss - so if output comes out weird, check the type first.

Inputs and outputs

• clip_name1 through clip_name4 - one file per slot, mix formats freely.
• type - architecture selector, same list as core.

Output is a single CLIP into a CLIP Text Encode. Nothing else to it.

Install

ComfyUI Manager - search "Smart Model Loaders" - or:

cd ComfyUI/custom_nodes
git clone https://github.com/Maxed-Out-99/ComfyUI-SmartModelLoaders-MXD
cd ComfyUI-SmartModelLoaders-MXD
pip install -r requirements.txt

Restart; it's under advanced/loaders. No model downloads, light requirements (gguf>=0.13.0, sentencepiece, protobuf).

When to skip it

If you don't have a four-encoder model, this node is not for you - the dual loader handles Flux/SDXL, the triple handles SD3, and those are the real workhorses of this pack. The quad is here for the person who knows they need it, and if that's you, it's a clean one-node solution to a problem nothing else in core ComfyUI addresses. Otherwise, file it under "good to know it exists."