GGUF TripleCLIP Loader

Most GGUF pipelines only need one or two text encoders wired up. This one's for the models that need three - think SD3-family setups that were designed around three separate encoders working together (CLIP-L, CLIP-G, and T5-XXL, in the general architecture this style of loader targets). If your model card explicitly says it needs three encoder files, this is the node; if it doesn't, you almost certainly want the plain or Dual CLIP loader instead.

What's different about this one

Unlike its single- and dual-encoder siblings, this node has no type selector - the schema just doesn't expose one. Practically, that means less to configure: you're not picking an architecture pairing, you're just handing it three files. It's a smaller surface to get wrong, at the cost of a bit less flexibility if you were hoping to reuse it across architectures the way you can with the type-driven loaders.

Inputs and outputs

Three required inputs, all the same kind: clip_name1, clip_name2, clip_name3 - pick a gguf or safetensors text-encoder file for each. Get the order right for whatever your specific model expects; triple-encoder setups typically care which file goes in which slot, even without a type dropdown to remind you.

One output: CLIP, which wires into CLIP Text Encode exactly like any other CLIP output - the fact that three files went into building it is invisible downstream.

Installing it

ComfyUI Manager: search gguf, install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/calcuis/gguf

No extra dependencies to install on the current version. All three encoder files go in ComfyUI/models/text_encoders - same folder as everything else in this family, gguf and safetensors side by side.

When you actually need this

This is genuinely the least-used node in the pack, and that tracks: three-encoder architectures are a smaller slice of what people run day to day compared to the one- or two-encoder models most workflows are built around. Don't reach for it out of curiosity - reach for it when a specific model's documentation tells you it needs three separate text-encoder files loaded together, and use the plain single-encoder loader or the Dual loader for everything else.

Common issues

Wrong file order. With no type selector to catch a mismatch, a scrambled clip order is easy to end up with and won't necessarily throw an error - you'll just get conditioning that doesn't line up with what the model expects. If output looks off, this is the first thing to check.

Cross-pack loader confusion. This is a different, separately-coded implementation from city96's more common ComfyUI-GGUF pack. The two read the same gguf format but aren't guaranteed to behave identically loading the same file - there are reports of a text-encoder file loading cleanly in one pack and throwing a hard type error (int64) when run through the other's loader instead. If a triple-encoder setup is failing partway, make sure you're not accidentally mixing loader implementations across packs for the same set of files.

Empty dropdowns. Files not in text_encoders, or added after ComfyUI started - restart to refresh the list.

No obvious node for your model. If a model needs a different encoder count than one, two, three, or four, none of this pack's CLIP loaders will fit - check whether the model's own recommended workflow uses a different node entirely.