Triple CLIP Selector

Some models don't have one text encoder, they have several. SD3 and Flux.1 want a CLIP plus a T5; SDXL wants two CLIPs of different sizes. Setting that up in stock ComfyUI means a DualCLIPLoader, some fiddling with type options, and squinting at dropdowns. Triple CLIP Selector exists to make the three-encoder case boring - which is the point of Sage Utils' whole selector family.

It's exactly what the name says: three dropdowns - clip_name_1, clip_name_2, clip_name_3 - each listing the CLIP/text-encoder files in your models/clip (and similar) folders, and one output. Nothing gets loaded when you use it. A selector just records which files you picked and packages them as a clip_info value of type MODEL_INFO, which the loader half of the family then consumes. The entire Sage Utils model philosophy is split like this: selectors pick and describe, loaders actually load. You're building a description of your model setup first, and only at execution does anything touch disk or VRAM.

That clip_info output is the thing you'll wire into the pack's Load CLIP node (officially "CLIP Loader From Info" under Sage Utils/loader), which expands into the right core loader with the right type baked in. That's the workflow shape: Triple CLIP Selector → Load CLIP → CLIP Text Encode → conditioning. If you also picked a UNET and a VAE with their selectors, you can bundle all three into a UNET CLIP VAE To Model Info node and treat them like a single virtual checkpoint - a pattern that pays off when you keep your model components separate (say, a distilled UNET plus a swapped-in VAE) but want them to flow through the metadata system like one thing.

Two things worth knowing before you grab it. First, this is the "three" variant - the pack also ships a Quad CLIP Selector if you're running something that genuinely wants four encoders. Second, "CLIP" here is doing loose duty for "text encoder." The file list it pulls comes from the pack's model registry, and newer architectures with LLM-style encoders (T5, Qwen, Chroma's T5 tokenizer options) are handled through the same slots - the tooltip even notes the Chroma variant exists for T5 tokenizer tweaks. If your model needs a specific encoder type, don't assume the first dropdown value is right; check the type your loader expects.

Install is the standard pack story: ComfyUI Manager (search "Sage Utils") or

cd ComfyUI/custom_nodes
git clone https://github.com/arcum42/ComfyUI_SageUtils.git

then restart. No models, no heavy deps - the selector itself is pure dropdown plumbing. If you installed the pack for its LLM sidebar features, the startup log may complain about unreachable LM Studio/Ollama endpoints; that's cosmetic and unrelated to this node.